CN101442673A - Method for encoding and decoding Bell formwork image - Google Patents

Abstract

The invention relates to digital image compression technology, in particular to a Bell template image encoding and decoding method. It solves the problem that the existing method of directly compressing the original Bell template image data is difficult to achieve a good balance between the compression ratio, the reconstructed color image quality and the complexity. The present invention adopts LOCO-I technology to perform lossless encoding on the green component; the green component at the position of the red and blue components is estimated by the gradient interpolation method, and the distance between the wavelet low frequency subband of the red and blue component and the estimated green component low frequency subband can be obtained Color-difference signal, this color-difference signal can be compressed losslessly or nearly losslessly by LOCO-I technology. At the decoding end, the lossless green component must be obtained first, and the green component at the position of the red and blue components is estimated as in the encoding process, and the wavelet high frequency subband of the estimated green component can replace the corresponding wavelet high frequency subband of the red and blue component , and the rest is the inverse process of encoding. The invention has the characteristics of high efficiency and low complexity.

Description

Bell template image encoding and decoding method

technical field

The invention relates to a digital image processing technology, in particular to a digital image compression technology, specifically a Bell template image encoding and decoding method.

Background technique

With the wide application of image sensors similar to Bell color filter array (referred to as Bell template) in digital cameras and video cameras, in order to improve the storage and transmission efficiency of digital images, image data compression technology is becoming more and more important.

In traditional digital camera products, the digital image with similar Bell template format (referred to as Bell template image) is interpolated to estimate the other two missing color values at each pixel position, thereby obtaining full-color image data, and then Use JPEG2000 or JPEG international standard based on wavelet transform to compress the interpolated full-color data. Obviously, this traditional method produces new data redundancy in the interpolation process, thus reducing the compression efficiency. In recent years, direct compression of raw Bell template image data has become a new research direction, resulting in many excellent lossless or lossy compression methods. Compared with the traditional compression method, the advantage of directly compressing the Bell template image is that the processor of the digital camera or video camera does not need to execute the interpolation algorithm, which greatly reduces the calculation load. 1/1, which is beneficial to improve the coding efficiency. Bell template image decoding and interpolation reconstruction of full-color images can be completed offline on a PC.

At present, the compression ratio provided by the Bell template image lossless compression algorithm is very low, which cannot meet the needs of many practical applications. In order to improve the compression ratio, a variety of lossy compression algorithms based on international standards such as JPEG2000 have been produced. The disadvantages of these algorithms are high complexity, and each component in the decompressed Bell template image is distorted and has a high degree of distortion. As a result, even at a low compression rate, it is difficult to reconstruct a satisfactory full-color image with an excellent interpolation algorithm. In short, these algorithms are difficult to achieve a good balance between compression ratio, reconstructed color image quality and complexity.

Contents of the invention

The present invention provides a Bell template image encoding and decoding method to solve the problem that it is difficult to achieve a good balance between the compression ratio, the reconstructed color image quality and the complexity in the existing method of directly compressing the original Bell template image data. This method implements lossy compression and decompression of Bell template images. While obtaining a higher compression ratio, this method ensures that the quality of the reconstructed color image is closer to that of the reconstructed color image based on the Bell template image based on lossless codec, and takes into account the low complexity of the algorithm.

The present invention is realized by adopting the following technical solutions: Bell template image encoding and decoding method, including encoding and decoding process, the encoding process comprises the following steps: 1, separate the red, green and blue components in the Bell template image, and separate the red, blue The color component becomes a compact rectangular image, and the green component remains a quincunx-shaped image; 2. Losslessly compress the green component data, the first step: use the causal interpolation method to estimate each A green pixel value missing at the left and upper pixel positions of the original green pixel currently to be encoded (generally, the first two rows and two columns of the green quincunx image are used as encoding buffers, which are not processed), so as to meet LOCO-I The requirements of the encoder predictor and context modeling, the estimated value of the green pixel at the pixel position to the left of the original green pixel currently to be encoded is equal to the average of the original green pixel values to the left and above the pixel position, each currently to be encoded The estimated value of the green pixel at the pixel position above the original green pixel is equal to the average value of the original green pixel values to the left and right of the pixel position and the original green pixel value above the pixel position and divided by 2, where the encoding is performed At the end of each line of the green quincunx image, the estimated value of the green pixel at the pixel position above the original green pixel to be encoded is equal to the average value of the original green pixel values on the left and above the pixel position (obviously, this method uses the real original The green pixel value is interpolated to estimate the missing green pixel value, which meets the raster order encoding requirements of the LOCO-I encoder and effectively reduces the coding complexity of the green channel); the second step: apply the LOCO-I encoder to each original green Pixel encoding (that is, only the original green pixel encoding in the Bell template image, the estimated green value is not encoded); 3, using gradient interpolation (gradient interpolation is the prior art) to estimate the original red and blue in the Bell template image The green pixel value missing at the pixel position, and separate the estimated green pixel at the original red and blue pixel positions, and turn it into a green compact rectangular image corresponding to the red and blue compact rectangular images in step 1; 4. For step The red and blue compact rectangular images obtained in step 1 and step 3 and their corresponding green compact rectangular images are respectively subjected to wavelet transformation; Subtract, obtain red-green low-frequency sub-band color difference and blue-green low-frequency sub-band color difference; 6, apply LOCO-I coder to red-green low-frequency sub-band color difference and blue-green low-frequency sub-band color difference to carry out nearly lossless or lossless coding; 7. Combine the code stream obtained after encoding the original green component obtained in step 2 and step 6 with the code stream obtained after the red-blue low-frequency sub-band color difference encoding to form a complete Bell template image compression code stream, and complete the encoding process; the decoding process Including the following steps: 1. Obtain the green component code stream of the Bell template image from the composite code stream, apply the LOCO-I decoder to obtain the original green pixel values, and arrange them according to the quincunx image; before decoding each original green pixel, it is necessary to use The causal interpolation method obtains the left and upper positions of the current original green pixel to be encoded This is the same as the causal interpolation method in step 2 in the encoding process; 2. Use the gradient interpolation method to estimate the green pixel value missing from the original red and blue pixel positions, and convert the original red and blue pixels to The green pixels estimated at the color pixel positions are separated and become green compact rectangular images corresponding to the red and blue components respectively, which is the same as step 3 in the encoding process; 3. Perform wavelet transformation on the green compact rectangular images corresponding to the red and blue components , obtain four sub-bands respectively; 4, obtain red-green, blue-green low-frequency sub-band color-difference code streams from composite code stream, apply LOCO-I decoder to obtain red-green, blue-green low-frequency sub-band color-difference signals; 5 1. Decoding the red and blue component data in the Bell template image, the process of decoding the red component data is as follows: the first step: the red-green low-frequency sub-band color difference signal obtained in step 4 and the low-frequency corresponding green compact rectangular image obtained in step 3 The sub-bands are subtracted to obtain the low-frequency sub-band of the red component. The second step: the high-frequency sub-band of the green compact rectangular image corresponding to the red component obtained in step 3 is regarded as the high-frequency sub-band of the red component. The third step: so far Obtain the four wavelet subbands of the red component, and restore the red compact rectangular image through wavelet inverse transformation; same as the process of decoding the red component data, the blue component can be decoded at the same time to restore the blue compact rectangular image; 6. Obtained in step 1 Based on the original quincunx-shaped green component image, according to the position of the original red and blue pixels in the Bell template image before encoding, arrange each pixel in the decoded red and blue compact rectangular image, and restore it into a Bell template image.

The LOCO-I codec described in this invention is prior art. The LOCO-I codec is shown in the document "Marcelo J.Weinberger, Gadiel Seroussi, Guillermo Sapiro," The LOCO—Ilossless image compression algorithm: principles and standardization into JPEG—LS" [J]. IEEE Trans Image Processing, 2000, 9 (6): 1309-1364".

The invention separates the Bell template image into three-color component channels, performs lossless encoding and decoding on the green component based on causal interpolation, retains the main brightness information of the original color image, effectively utilizes the wavelet subband correlation of the color component, and converts the red and blue components Perform lossy codecs. In order to reduce the complexity of the algorithm, the codecs of the three channels all adopt the low-complexity LOCO-I predictive difference codec technology.

The LOCO-I technology based on causal interpolation is used to encode the green component losslessly; the green component at the position of the red (blue) component is estimated by the gradient interpolation method, and the corresponding wavelet height between the estimated green component and the red (blue) component There is a strong correlation between the frequency subband information, that is, the estimated green component wavelet high frequency subband can replace the corresponding wavelet high frequency subband of the red (blue) component, and further the red (blue) component low frequency subband and the estimated Estimate the color-difference signal between the low-frequency sub-bands of the green component. This color-difference signal can be lossless or near-lossless compressed by the low-complexity LOCO-I technology, thereby completing high-performance Bell template image coding. At the decoding end, the lossless green component must be obtained first, and the rest is the inverse process of encoding. Experiments show that the average PSNR of the Bell template image after decoding is more than 42dB compared with the Bell template image before encoding, and the visually lossless full-color image can be obtained by applying the high-performance adaptive filter interpolation method.

The encoder obtains three-way components after the color components are separated, that is, the red and blue components are compacted into rectangular images, and the green component maintains a quincunx-shaped image, and each component image is encoded separately, and finally the multiple code streams are combined into the final compressed code stream . In the Bell template image, the green component accounts for one-half of the total number of pixels and contains the main brightness information of the original image. Therefore, for the green component, LOCO-I technology is used for lossless coding. Before encoding each raw Bell template pixel, a causal interpolation method should be used to estimate the missing green value at the red and blue pixel positions to obtain a full-resolution green component image, which provides a match for the LOCO-I predictor and context model. data. It should be noted that the present invention only encodes the green pixels in the original Bell template image, and the green values estimated at the red and blue pixel positions are not encoded. When the Bell template red component is encoded, the compact rectangular red component and the rectangular green component interpolated and estimated at the same position are subjected to one-layer wavelet transformation. The preferred bior3.3 wavelet of the present invention, experiments show that the corresponding wavelet high-frequency subbands There is a strong correlation between. At the decoding end, the green component of the original Bell template image can be decoded losslessly, and the rectangular green component at the same position as the original Bell template red component can be estimated by interpolation. In the present invention, the wavelet high-frequency sub-band of the rectangular green component is regarded as the high-frequency sub-band of the original Bell template red component, so the high-frequency sub-band of the Bell template red component does not need to be encoded and transmitted at the encoding end, only the Bell template needs to be encoded The low-frequency subband of the red component has only a quarter of the data of the red component of the original Bell template. Further, the low-frequency subband is subtracted from the low-frequency subband of the estimated rectangular green component to obtain the low-frequency sub-band color difference signal, which is effective Redundancy between red and green components is effectively removed, and the complexity of coding is reduced. Finally, LOCO-I technology is used for lossless or near-lossless (δ=1, δ=2) coding. Similarly, the blue component of the Bell template can be encoded.

At the decoding end, firstly, the LOCO-I decoder based on causal interpolation is used to decode the green component losslessly to obtain the original Bell template quincunx image. It should be noted that, just as in the LOCO-I encoder, for each original Bell template green pixel to be decoded, the predictor and context model in its decoder also need to obtain the missing green estimate at the past red or blue pixel position , the corresponding estimation method still adopts the causal interpolation method in the encoder. Decode the Bell red component on the basis of the Bell template quincunx-shaped green component image, use the same gradient interpolation method as the encoding end to estimate the green pixel value at the red pixel position of the original Bell template image, and compact it into a green rectangular image, and then proceed the same as the encoding end The wavelet transform of , get a green low-frequency sub-band and three green high-frequency sub-bands. While the green component of the Bell image is losslessly decoded, the low-frequency sub-band of the difference between red and green is also decoded, and subtracted from the green low-frequency sub-band obtained just now to obtain the low-frequency sub-band of the red component, and finally three green high-frequency sub-bands have been obtained Band synthesis and inverse wavelet transform result in a compact rectangular red component, which is obviously distorted compared with the original Bell red component. In the same way, the original Bell blue component can be decoded. Finally, the decoded rectangular red and blue components and quincunx-shaped green components can be restored to the Bell image according to the positions in the original Bell template. Experiments show that the peak signal-to-noise ratio between the decoded Bell template image and the original Bell template image ( PSNR) can reach above 42dB on average.

Since the green component of the Bell template image contains the main luminance information in the original color image, the present invention performs lossless compression on the green component and lossy compression on the red and blue components, which ensures that the present invention has the same compression rate as the above-mentioned Compared with the lossy compression method, the quality of the reconstructed color image is closer to the quality of the color image reconstructed based on the lossless codec Bell template image, and has a lower complexity.

Figure 8 shows the process of evaluating codec performance with PSNR. First, the original color image is down-sampled to obtain the original Bell template image, and the Bell template image obtained after encoding and decoding is adaptively filtered and interpolated to obtain a reconstructed color image, and the peak signal-to-noise between the original color image and the reconstructed color image is calculated Ratio (PSNR), the specific calculation formula is as follows

$\mathrm{<span\; class="notranslate">\; PSNR</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 10</span>}\mathrm{<span\; class="notranslate">\; lg</span>}<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; 255</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; W</span>}}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; W</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span>$

Among them, I ₁ represents the original color image data, I ₂ represents the color image data reconstructed by interpolation after decoding, x, y represent the coordinates of pixel points, H, W represent the effective height and width of the image.

The adaptive filter interpolation described in the literature "Dmitriy Paliy, Vladimir Katkovnik, RaduBilcu, Sakari Alenius, Karen Egiazarian, "Spatially Adaptive Color Filter Array Interpolation for Noiseless and Noisy Data", Wiley Periodicals, Inc.Vol.17, 105-122 ( 2007)"

Figure 9 is 24 color images (24bit/pixel) selected from the Kodak standard image set, which are widely used in experiments of various algorithms for processing Bell template images. Number each picture from 1-24 in order of arrangement from left to right and top to bottom.

Technical effect of the present invention is as follows:

1) An efficient and low-complexity Bell template image data compression and decompression method based on wavelet transform is provided.

2) The algorithm structure of the present invention is simple, the amount of calculation is small, and the integer operation is completely adopted, and the wavelet transform and the LOCO-I codec can be reused in each codec channel, which occupies less hardware resources and is easy to implement in hardware.

3) Use the flow chart shown in Figure 8 to evaluate the effect of the present invention. 24 full-color images (Fig. 9) from the Kodak standard picture collection are selected as the experimental data. The PSNR values produced by the present invention and JPEG2000 are listed in Table 1 at a given bit rate. Experimental results show that under the three bit rates, the PSNR values obtained for almost all test images are higher than the corresponding values of JPEG2000, and are closer to the results of direct interpolation of the original Bell template, and the reconstructed color images achieve visually lossless effects. The complexity of the present invention is significantly lower than that of JPEG2000.

In Table 1, the second column represents the PSNR value of the color image obtained after direct interpolation and reconstruction of the original Bell template image using the adaptive filter interpolation method, and the second column is obtained by interpolation and reconstruction after decoding under three code rates in the present invention. The PSNR value of the color image, the third column is the PSNR value of the color image obtained by interpolation and reconstruction after JPEG2000 decoding at three bit rates. bpp represents the compression rate, that is, the average number of bits per pixel. Wherein 2.73bbp is the compression ratio of the color difference signal lossless coding (δ=0), 2.56bbp is the compression ratio of the lossy coding method (δ=1), and 2.49bbp is the compression ratio of the lossy coding (δ=2).

Table 1 The effect of the present invention is compared with the PSNR of JPEG2000

Description of drawings

Fig. 1 is the flowchart of encoding process;

Fig. 2 is the flowchart of decoding process;

Figure 3 Bell template image color component separation diagram;

Figure 4 Schematic diagram of green component causal interpolation;

Figure 5 is a schematic diagram of green component gradient interpolation;

Fig. 6 flow chart of red component encoding channel;

Figure 7 is a flow chart of the red component decoding channel;

Figure 8 PSNR evaluation schematic;

Figure 9 Kodak standard image set;

Detailed ways

Bell template image encoding and decoding method, including encoding and decoding process, the encoding process comprises the following steps: 1, separate the red, green and blue components in the Bell template image, and turn the separated red and blue components into a compact rectangular image, and the green component Keep it as a quincunx-shaped image; 2. Losslessly compress the green component data, the first step: use the method of causal interpolation to estimate the left side and For the missing green pixel value at the upper pixel position, the estimated value of the green pixel at the left pixel position of each current original green pixel to be encoded is equal to the average value of the original green pixel values at the left and upper of the pixel position, and each current original green pixel to be encoded The green pixel estimate for a pixel position above the pixel is equal to the average of the original green pixel values to the left and right of that pixel position added to the original green pixel value above that pixel position and divided by 2, where encoding proceeds to green At the end of each line of the quincunx image, the estimated value of the green pixel at the pixel position above the original green pixel to be encoded is equal to the average value of the original green pixel values to the left and above the pixel position; the second step: apply the LOCO-I encoder Encode each original green pixel; 3. Use the gradient interpolation method to estimate the missing green pixel values at the original red and blue pixel positions in the Bell template image, and separate the green pixels estimated at the original red and blue pixel positions to become Become a green compact rectangular image corresponding to the red and blue compact rectangular images in step 1; 4, carry out wavelet transform respectively to the red and blue compact rectangular images and their corresponding green compact rectangular images obtained in steps 1 and 3; 5 1. After wavelet transform, the red-blue compact rectangular image low-frequency subband is subtracted from its corresponding green compact rectangular image low-frequency sub-band to obtain red-green low-frequency sub-band color difference and blue-green low-frequency sub-band color difference; 6. Apply LOCO-I encoding The red-green low-frequency sub-band color difference and the blue-green low-frequency sub-band color difference are nearly lossless or lossless encoded; 7. The code stream and the red-blue low-frequency sub-band color difference obtained after encoding the original green component obtained in steps 2 and 6 The code stream obtained after encoding forms a complete Bell template image compression code stream, and completes the encoding process; the decoding process includes the following steps: 1. Obtain the Bell template image green component code stream from the composite code stream, and apply the LOCO-I decoder Obtain the original green pixel value and arrange it according to the quincunx image; before decoding each original green pixel, it is necessary to use the causal interpolation method to obtain the estimated value of the green pixel at the left and upper positions of the original green pixel to be encoded, which is the same as that in the encoding process The causal interpolation method in step 2 is the same; 2. Use the gradient interpolation method to estimate the missing green pixel values at the original red and blue pixel positions, and separate the estimated green pixels at the original red and blue pixel positions, respectively Become a green compact rectangular image corresponding to the red and blue components, which is the same as step 3 in the encoding process; 3. Carry out wavelet transformation to the green compact rectangular image corresponding to the red and blue components to obtain four subbands respectively; 4. From the composite code Obtain the red-green, blue-green low-frequency sub-band color difference code stream in the stream, and apply LO The CO-I decoder obtains red-green, blue-green low-frequency sub-band color difference signals; 5, decodes the red and blue component data in the Bell template image, and the process of decoding the red component data is as follows: the first step: the obtained in step 4 The red-green low-frequency sub-band color difference signal is subtracted from the low-frequency sub-band corresponding to the green compact rectangular image obtained in step 3 to obtain the low-frequency sub-band of the red component. The second step: the green compact rectangular image corresponding to the red component obtained in step 3 The high-frequency sub-band is regarded as the high-frequency sub-band of the red component, the third step: the four wavelet sub-bands of the red component have been obtained so far, and the red compact rectangular image is restored by wavelet inverse transformation; the same as the process of decoding the red component data, The blue component can be decoded at the same time to restore the blue compact rectangular image; 6. On the basis of the original quincunx-shaped green component image obtained in step 1, arrange the decoded image according to the position of the original red and blue pixels in the Bell template image before encoding Each pixel in the red-blue compact rectangular image is restored to a Bell template image. The wavelet transform in the present invention selects bior3.3 wavelet transform, and the wavelet inverse transform selects bior3.3 wavelet inverse transform to further improve the efficiency of encoding and decoding.

The present invention will be further described in conjunction with accompanying drawing:

Fig. 1 is a flowchart of the encoding process of the present invention. First, the color components of the Bell template image are separated, the green component remains in the shape of a quincunx, and the red and blue components become a compact rectangular image. The encoding is divided into three channels, and each channel is encoded by a low-complexity LOCO-I encoder, which reduces the algorithmic complexity of the encoder as a whole. Since the green component contains the main information of the brightness of the Bell template image, and also provides information for the coding channel of the red and blue components, the green component is losslessly compressed. In the green encoding channel, the function of the causal interpolation module is to ensure that the missing green pixel value in the corresponding context is estimated before encoding a certain green pixel value. The switch K1 indicates that only the codeword corresponding to the original green pixel in the quincunx-shaped image is output, and Codewords for green pixels estimated by causal interpolation are not output. In the red component encoding channel, gradient interpolation is performed based on the green quincunx image to estimate the green pixel value of the red component pixel position, and then a layer of wavelet decomposition is performed on the red component compact image and the green compact image at the same pixel position respectively. The wavelet is bior3.3, and the corresponding coefficients of the two low-frequency sub-bands are subtracted to obtain the red-green low-frequency sub-band color difference signal, and finally the LOCO-I encoder is used for lossless or near-lossless encoding. The blue channel can be encoded in the same way. Switch K2 indicates that the red and blue encoding channels can be multiplexed with the LOCO-I encoder in sequence. Finally, the code streams of the three channels are combined into a complete Bell template image code stream.

Fig. 2 is a flowchart of the decoding process of the present invention. Corresponding to the encoder, the decoder is also divided into three channels, which decode the red, green and blue components respectively, and the red and blue decoding channels can work in parallel. First, the code stream of each component is separated from the code stream, and then the LOCO-I decoder and the causal interpolation module are applied to obtain the original green pixels, and these pixels are arranged into a quincunx-shaped image at the code end. Then apply the gradient interpolation method at the encoding end to estimate the green pixel value of the original red and blue pixel positions to form their corresponding green compact rectangular images, and perform a layer of wavelet decomposition to obtain their respective four wavelet subbands. When decoding the red channel, first apply the LOCO-I decoder to obtain the red-green low-frequency sub-band color difference signal, and then subtract it from the low-frequency sub-band of the corresponding green rectangular image to obtain the red component low-frequency sub-band, the low-frequency sub-band and the red component The high-frequency sub-bands of the corresponding green rectangular image are synthesized into four sub-bands and subjected to wavelet inverse transformation to obtain the decoded value of the original red component. In the same way, blue channel decoding can be performed. Switch K2 indicates that the red and blue coding channels can be multiplexed with the LOCO-I decoder in turn. Finally, the decoded pixels of each channel are recombined according to the position of the original Bell template image, and the Bell template image is restored.

Fig. 3 is a schematic diagram of separating color components of a Bell template image. After separation, the green component remains a quincunx image, and the red and blue components become compact rectangular images.

Figure 4 is a schematic diagram of causal interpolation. Based on the green quincunx-shaped image separated in Figure 3, assuming that the pixel to be encoded is G ₃₄ , the LOCO-I encoder needs to know the missing green pixel values on the left and upper sides of G ₃₄ , which are denoted by G _L and G _U respectively. The estimation method as follows

In addition, if the pixel to be encoded reaches the end of the image line, let the pixel to be encoded be G ₃₆ , the estimation method of _GL remains unchanged, and the estimation method of G _U becomes:

Obviously, this method only uses the past real pixels of the pixel to be encoded to interpolate and estimate the missing green pixel value, which meets the raster order encoding requirements of the LOCO-I encoder and effectively reduces the encoding complexity of the green channel.

为例，方法如下Fig. 5 is a schematic diagram of the gradient interpolation of the green component. Based on the green quincunx image separated in Figure 3, estimate the value of the green rectangular image corresponding to the red and blue component rectangular image to estimate the missing green pixel at the R ₃₃ pixel position

For example, the method is as follows

${\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; G</span>}}}_{\mathrm{<span\; class="notranslate">\; 33</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; round</span>}<span\; class="notranslate">\; (</span>\frac{<span\; class="notranslate">\; \&dtri;</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; \&dtri;</span>\mathrm{<span\; class="notranslate">\; V</span>}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}}{<span\; class="notranslate">\; \&dtri;</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; \&dtri;</span>\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; )</span>$

in:

$G_h=\frac{1}{2}(G_{32}+G_{34}),$ $G_V=\frac{1}{2}(G_{\mathrm{twenty\; three}}+G_{43})$

$\&dtri;h=|G_{32}-G_{34}|,$ $\&dtri;V=|G_{\mathrm{twenty\; three}}-G_{43}|$

In the same way, the green pixel values missing in other positions can be estimated. Finally, the estimated green pixels are arranged into a green rectangular image corresponding to the red and blue component rectangular image in Fig. 3 .

和R_LL，令

和R_LL子带中对应系数相减，得到红-绿低频子带色差信号

，并输入LOCO-I编码器，可以选择无损编码(δ＝0)或有损编码方式(δ＝1，δ＝2)。蓝分量编码通道工作原理与此相同。FIG. 6 is a flow chart of the red component encoding channel. The red component rectangular image and its corresponding green rectangular image are decomposed by one layer of bior3.3 wavelet to obtain their respective low-frequency subbands, which are expressed as

and R _LL , let

Subtract the corresponding coefficients in the R _LL sub-band to obtain the red-green low-frequency sub-band color difference signal

, and input to the LOCO-I encoder, lossless coding (δ=0) or lossy coding (δ=1, δ=2) can be selected. The blue component encoding pass works the same way.

，同时原始红分量图像对应的绿色矩形图像通过bior3.3小波一层分解后获得四个子带

和令

，则得到红分量低频子带R_LL，进一步与

合成四个小波子带并通过bior3.3小波逆变换，即可得到解码的红分量矩形图像。蓝分量解码通道工作原理与此相同。FIG. 7 is a flow chart of the red component decoding channel. LOCO-I decoder outputs red-green low frequency sub-band color difference signal

At the same time, the green rectangular image corresponding to the original red component image is decomposed by one layer of bior3.3 wavelet to obtain four subbands

and make

, then the red component low-frequency sub-band R _LL is obtained, further combined with

By synthesizing four wavelet subbands and inverse bior3.3 wavelet transformation, the decoded red component rectangular image can be obtained. The blue component decoding pass works the same way.

Claims (2)

1, a kind of Bell formwork image encoding and decoding method, comprise the Code And Decode process, it is characterized by: cataloged procedure comprises the steps: 1, separates RGB component in the Bell formwork image, and isolated redness, blue component become compact rectangular image, green component remains plum blossom shape image; 2, the lossless compress green component data, the first step: adopt the method for cause and effect interpolation to estimate to go except that preceding two of green plum blossom shape image, each the current original green pixel to be encoded left side that preceding two row are outer and the green pixel values of top location of pixels disappearance, the green pixel estimated value of each current original green pixel left pixel to be encoded position equals the mean value of the original green pixel values of this location of pixels left side and top, the green pixel estimated value of each current original green pixel to be encoded top location of pixels equals the original green pixel values addition of top of the mean value of original green pixel values on the left side of this location of pixels and right side and this location of pixels again divided by 2, wherein coding proceeds to the every row of green plum blossom shape image when terminal, and the green pixel estimated value of current original green pixel to be encoded top location of pixels equals the mean value of the original green pixel values of the left side of this location of pixels and top; Second step: use the LOCO-I encoder to each original green pixel coding; 3, adopt the gradient interpolation method to estimate the green pixel values of original redness, blue pixel topagnosis in the Bell formwork image, and the green pixel of estimating on original redness, the blue pixel position separated, become with step 1 in redness, the green compact rectangular image that blue compact rectangular image is corresponding; 4, compact rectangular image of red indigo plant and the corresponding green compact rectangular image thereof that obtains in step 1 and the step 3 carried out wavelet transformation respectively; 5, behind the wavelet transformation, the green compact rectangular image low frequency sub-band that the compact rectangular image low frequency sub-band of red indigo plant is corresponding with it subtracts each other, and obtains red-green low frequency sub-band aberration and indigo plant-green low frequency sub-band aberration; 6, use the LOCO-I encoder red-green low frequency sub-band aberration and indigo plant-green low frequency sub-band aberration are carried out closely harmless or lossless coding; 7, the bit stream complex that obtains behind code stream that obtains behind the original green component coding of step 2 and step 6 acquisition and the red blue low frequency sub-band aberration coding is formed complete Bell formwork image compressed bit stream, and finish cataloged procedure;

Decode procedure comprises the steps: 1, obtains Bell formwork image green component code stream from composite bit stream, uses the LOCO-I decoder and obtains original green pixel values, and arrange according to plum blossom shape image; 2, adopt the gradient interpolation method to estimate the green pixel values of original redness, blue pixel topagnosis, and the green pixel of estimating on original redness, the blue pixel position is separated, become the green compact rectangular image corresponding respectively with red blue component; 3, the green compact rectangular image to red blue component correspondence carries out wavelet transformation, obtains four subbands respectively; 4, from composite bit stream, obtain red-green, indigo plant-green low frequency sub-band aberration code stream, use the LOCO-I decoder and obtain red-green, indigo plant-green low frequency sub-band color difference signal; 5, red, the blue component data in the decoding Bell formwork image, the process of red component data of decoding is as follows: the first step: red-green low frequency sub-band color difference signal that step 4 is obtained subtracts each other with the low frequency sub-band of the corresponding green compact rectangular image that step 3 obtains, obtain the low frequency sub-band of red component, second step: the high-frequency sub-band that step 3 is obtained the green compact rectangular image of corresponding red component is considered as the high-frequency sub-band of red component, the 3rd step: so far obtained four wavelet sub-bands of red component, recovered red compact rectangular image through wavelet inverse transformation; The same with the process of the red component data of decoding, can decode blue component simultaneously and recover blue compact rectangular image; 6, on the basis of the original plum blossom shape green component image that step 1 obtains, according to the position of original red blue pixel in Bell formwork image before the coding, each pixel in the compact rectangular image of red indigo plant behind the permutation decoding reverts to Bell formwork image.

2, Bell formwork image encoding and decoding method as claimed in claim 1 is characterized by: wavelet transformation is selected the bior3.3 wavelet transformation, and wavelet inverse transformation is selected the bior3.3 wavelet inverse transformation.

Images