KR100192780B1 - Contour encoder - Google Patents

Abstract

The present invention relates to a contour encoder using an adaptive ratio (B) -spline, comprising: a segmentation unit 41 for segmenting an input image; A feature point extracting unit 42 for extracting feature points from the segmented image; A ratio (B) -spline portion (43) for obtaining a ratio (B) -spline by applying a polygonal approximation to the extracted feature point intervals up to a predetermined order; Error calculation is performed by calculating the difference between the applied ratio (B) spline and the input image for each feature point section, and selecting the ratio (B) spline having the minimum error by comparing the calculated errors. (44); And a vector quantizer 45 (VQ) for outputting encoded contour information by vector quantizing the ratio (B) -spline selected by the error calculation / comparison unit (44) to apply a fixed ratio (B) -spline. The image quality can be improved as compared with the conventional method.

Description

Contour encoder

1 is a block diagram illustrating a general object-oriented encoder.

2 is a diagram showing an example of polygonal approximation for extracting the contour of an image.

3 is a diagram showing an error vector of an outline.

4 is a block diagram illustrating a contour encoder using an adaptive ratio (B) -spline according to the present invention.

* Explanation of symbols for main parts of the drawings

11 image analysis unit 12 parameter encoder

13: parameter decoder 14: parameter memory

15: image synthesizer 21: the original object image

22: non-spline 41: segmentation unit

42: feature point extraction section 43: non-spline section

44: error calculation / comparison unit 45: vector quantization unit

V1 to V7: feature points (vertex)

The present invention relates to an object-oriented encoder, and more particularly to a contour encoder using an adaptive ratio (B) -spline.

In general, digital image coding and technology are widely known as transform coding, vector quantization, subband conding, prediction coding, and entropy coding. Efforts have been made to study the coding techniques while obtaining a standard for encoding video and video to obtain more improved coding techniques.

On the other hand, in the conventional coding method which processes the unit of block or macro block by using the mathematical statistical characteristics of the video signal, the lower the bit rate, that is, the higher the compression rate, the more blocking artifacts, mosquito artifacts and boundary flows. Problems such as blurring become serious and visually annoying.

Therefore, in order to reconstruct an image that satisfies the human vision, which is the subject of image quality evaluation, in a system requiring high compression ratio, the human visual system (HVS), which studies what a person sees and feels, actively uses an image signal. It is preferable to encode. In particular, the second-generation coding schemes such as object-based coding, model-based coding, segmentation-based coding, and fractal coding are actively researched as the international standardization of MPEG4 begins. It is becoming.

Object-oriented coding as a second generation coding technique for transmitting a video at ultra low speed, unlike conventional block-oriented hybrid coders, is not in the form of a square block. Encode by object in form. By transmitting not only the motion information and the signal information but also the shape information of the object, it is possible to prevent the deterioration of the reproduced picture quality such as the block phenomenon and the spot phenomenon which may appear in the block-oriented encoding. In addition, in the image analysis process, an image that is important to human vision, such as a detailed part of the face, can be reconstructed with higher image quality than block-oriented coding by transmitting more data than other regions.

In particular, object-oriented coding allows only a few objects to move at remarkable and moderate speeds, moving objects occupy 40 to 60% of the image, and the background is most efficient in still images. Therefore, object-oriented coding research is mainly used in application fields such as video telephone and video conferencing.

However, in object-oriented coding that performs encoding for each object, the contour information of the image must be sent, so an efficient contour coding method is required. Chain coding is the best known method of encoding an outline. However, chain coding, which is a lossless coding, has a large amount of data and thus is not suitable for ultra low-speed video encoding. Therefore, the contour must be transmitted by approximating the contour. Among the methods of approximating the contour, it is known that the polygon approximation is most suitable for the encoding method for each object. There is a downside to being unnatural. In addition, the conventional outline encoder using a spline has a significant deterioration in image quality because it uses a single B-spline or polygonal approximation.

Accordingly, the present invention has been made to solve the above-mentioned conventional problems, and after each non-B spline connection is performed from the first to nth order between each feature point (vertex) of the image, each obtained ratio (B) is obtained. To provide a contour encoder using an adaptive non-spline that can obtain the difference between the spline and the original image and improve the reproduced image quality by applying the non-spline with the smallest difference to the corresponding feature points. There is a purpose.

The contour encoder of the present invention for achieving the above object is a segmentation unit for segmenting an input image, a feature point extraction unit for extracting a feature point from the segmented image, by applying a polygon approximation to a predetermined order to the extracted feature point intervals A ratio (B) -spline unit for obtaining a ratio (B) -spline, calculates an error by calculating a difference between the applied ratio (B) -spline and an input image for each feature point section, and calculates an error and compares the calculated errors An error calculation / comparison for selecting a ratio (B) -spline having a vector, and a vector quantization unit for outputting hatched contour information by vector quantizing the ratio (B) -spline selected in the error calculation / comparison. .

As described above, the present invention can improve image quality compared to the conventional method of applying a fixed non-B spline by obtaining a non-B spline having a minimum error for each feature point section and transmitting the vector quantized. It works.

Hereinafter, with reference to the accompanying drawings will be described in detail the present invention.

First, a brief description of B-spline is provided to facilitate understanding of the present invention.

In displaying an outline which is a boundary of an object in an input image in two-dimensional space, a polygonal approximation is obtained after obtaining a vertex from the outline image 21 of the object as shown in FIG. It is a method of expressing the outline by several straight lines by fitting a straight line based on the vertex). At this time, the maximum distance d _max (also called an error) between the actual contour 21 and the approximate polygon 22 is used as a criterion for determining the degree of approximation, and the performance of the polygon approximation depends on this d _max .

That is, in approximating a curved outline, as shown in FIG. 2, vertices may be connected by lines, but adjacent vertices may be connected, or neighboring vertices may be connected by one or more, and two or more n may be connected. You can also connect neighboring vertices by crossing. A line passing between polygons in this figure is called a spline. Generally, an interpolation figure formed by grouping n neighboring contour points and connecting both ends thereof is called a convex hull. Then, the curve of the contour exists in this figure. The contour approximation obtained by grouping n neighboring contour points and connecting their ends is called n-dimensional B-spline.

Next, an object-based encoder to which the present invention is applied will be described.

As shown in FIG. 1, an object-based encoder applicable to the present invention may analyze an input image by referring to a source model and analyze the contour (M: shape, or contour), color (S: color), and motion (A :). motion) an image analyzer 11 for extracting parameters, a parameter encoder 12 for encoding the parameters (M, S, A) extracted by the image analyzer 11 with reference to a receiver model, and the encoded parameters A parameter decoder 13 for decoding (M, S, A), a parameter memory 14 for storing the decoded parameters (M, S, A), and the stored parameters (M, S, A) It consists of an image synthesizer 15 for synthesizing the previous image.

The basic concept of the object-oriented coder constructed as described above is to extract the moving objects and represent the motion (A: motion), contour (M: contour) and color (S: color) parameters. To express. In image analysis, which takes a sequence image as an input, each image is divided into moving object units. Each object i is represented by the actual parameter set of the motion parameter Ai, the contour parameter Mi, and the color parameter Si. These parameters depend on whether the source model you are using is two-dimensional or three-dimensional, and whether the properties of the object are to be viewed as rigid or non-rigid bodies. The calculated parameter set is parameter coded in the parameter encoder 12, which parameter coding depends on the receiver model which contains the assumptions about the visible sign and the error. The encoded parameters are sent to the receiving side, and are decoded by the parameter decoder 13 and stored in the parameter memory 14. The parameter memory of the encoder and the decoder can have the same parameter information so that the same image can be synthesized by the encoder / decoder. The displayed I _k ′ is used for encoding / decoding of the next image I _k-1 . Three parameters A, M, and S to be transmitted may be encoded in different ways to increase transmission efficiency. Typically, motion parameters are coded in DPCM, and contour parameters are encoded in a mixture of polygonal representations and spline representations. For objects that cannot be predicted by motion estimation, the object's color parameters are sent to the 8-bit PCM instead of the motion parameters. Therefore, three parameters are not always sent, but according to the characteristics of the moving object, as shown in Table 1 below.

This object-based encoding technique has the advantage of better object-preservation and excellent object-oriented prediction performance compared to H.261, MPEN-1, MPEG-2, etc., which process by block or macro block unit of fixed size.

In the object-oriented encoder as described above, the contour encoder using the adaptive non-spline of the present invention for encoding contour information by applying a plurality of orders of non-splines for each feature point section is shown in FIG. Segmentation unit 41 for segmenting the input image as shown in the; A feature point extractor 42 for extracting feature points from the segmented image; A ratio (B) -spline portion (43) for obtaining a ratio (B) -spline by applying a polygonal approximation to the extracted feature point intervals up to a predetermined order; Error calculation is performed by calculating the difference between the applied ratio (B) spline and the input image for each feature point section, and selecting the ratio (B) spline having the minimum error by comparing the calculated errors. (44); And a vector quantizer 45 (VQ) for outputting encoded contour information by vector quantizing the ratio (B) -spline selected by the error calculator / comparator 44.

Next, the operation of each component block will be described in detail.

The segmentation unit 41 of the present invention divides an input image signal. The segmentation unit 41 includes a region-based segmentation, a boundary-based segmentation, and a segmentation using a histogram. As a representative method of region-based segmentation, the split-and-merge method divides an image into small units and compares the similarities with neighboring pixels, sums them when they are determined to be the same area, and separates them if they are areas of different properties. The method of repeating and tracking the boundary line is a method of sequentially following the boundary line of the object using the slope value and the direction information of each pixel in the input image.

The feature point extractor 42 extracts a feature point from the segmented image, and it can be seen that there are seven feature points V1 to V7 in the image shown in FIG. ) Apply B-splines from the first to the nth order to the feature point intervals extracted by the feature point extractor 42, and the error calculation / comparator 44 is a non-spline as shown in FIG. After calculating the error of each interval, compare with each other to select the non-spline with the minimum error.

For example, as shown in FIG. 2, seven feature points (V1 to V7) are used for each section when the non-spline (B-spline) is applied to the fifth order (six sections in the embodiment of the present invention). Table 2 shows an example in which a low non-spline is selected.

As shown in Table 2, when non-spline encoding the image of FIG. 2, the third-order ratio (B) -spline is applied in the first section (V1-V2) and the second-order ratio in the second section (V2-V3). The (B) -spline is applied, and similarly, the third section is the third order, the fourth section is the third order, the fifth section is the first order, and the sixth section is the fourth order (B) -spline.

The vector quantization unit 45 according to the present invention transmits the contour information by vector quantization when the non-B spline having the minimum error is selected by the error calculation / comparison unit 44. Such beter quantization is widely known as the k-dimensional Euclidean space R. From R A mapping to a finite subset of, where finite set Y is called the codebook or codetable of the vector quantizer and It can be implemented by an encoder that gives one index to each input data X belonging to and a decoder that maps the index of transmission to a code vector belonging to Y.

As described above, the present invention improves image quality compared to the conventional method of applying a fixed non-B spline by obtaining a non-B spline having a minimum error for each feature point section and transmitting the vector quantized. It can be effected.

Claims (1)

A segmentation unit 41 for segmenting the input image; A feature point extractor 42 for extracting feature points from the segmented image; A ratio (B) -spline portion (43) for obtaining a ratio (B) -spline by applying a polygonal approximation to the extracted feature point intervals up to a predetermined order; Error calculation is performed by calculating the difference between the applied ratio (B) spline and the input image for each feature point section, and selecting the ratio (B) spline having the minimum error by comparing the calculated errors. (44); And a vector quantizer (V: VQ) for outputting encoded contour information by vector quantizing a ratio (B) -spline selected by the error calculation / comparison unit (44).