CN103268499B - Human body skin detection method based on multispectral imaging - Google Patents

Abstract

In the multi-spectral imaging system, the calibration method is used to obtain the reflection characteristics of the object under the two selected wave bands, and the reflection characteristics are used as features for skin detection. The multi-spectral imaging system includes a reference object with known reflection characteristics, a light source and a receiving device configured to capture images in two selected different wavebands. The acquisition of object reflection characteristics through calibration is to use the reflection characteristics of the reference object to predict the reflection characteristics of other objects in the image. The skin detection method uses the different reflection characteristics of human skin and non-skin objects as features, regards the detection of skin and non-skin as a binary classification problem, uses machine learning to train a classifier, and performs detection. <!--1-->

Description

Human skin detection method based on multispectral imaging

2. Technical field:

The invention relates to the fields of computer vision and image processing, in particular to a method for using computer vision to detect human skin. More specifically, the present invention relates to a low-cost and real-time multi-spectral imaging system to acquire images and obtain reflection characteristics of objects in the images through system and method design. At the same time, the invention relates to how to detect human skin under the complex background with dummies and masks.

3. Background technology:

Skin detection is an important research direction in the field of computer vision. For example, face detection and tracking in video conferencing, body language understanding, and instruction analysis in human-computer interaction all require precise skin detection algorithms. Most of the traditional skin detection algorithms are based on the color information obtained by the RGB camera for human skin color detection. The principle of the RGB camera is to synthesize color images through the three color channel components of R (red), G (green), and B (blue). Therefore, the most common The skin detection algorithm achieves the detection purpose by defining the value range of the skin color on each component in the color space. In this way, when the color value of a pixel falls within the defined interval, the pixel is considered to be skin. Regarding the choice of color space, Kakumanua et al. gave a good summary. However, when the detection crowd includes people from different races, or the background color is complex, and the external light is strong, the accuracy of color-based skin detection is relatively low. In addition, although the color space with illumination invariance is used, the influence of external illumination on the algorithm can be reduced to a certain extent. However, when the external light changes drastically, the performance of skin detection is still significantly reduced. Moreover, the skin detection algorithm based on RGB three-color imaging cannot effectively detect and remove items that are similar to human skin color, especially cannot distinguish between real people and dummies (including mannequins, masks, photos, etc.). Therefore, the skin detection algorithm based on RGB three-color imaging has great limitations.

The reflectance of human skin has been confirmed by physiology and experiments to have its unique characteristics: the reflectance is low in the ultraviolet part, and then increases with the increase of wavelength, and has a "W" shape at 575nm. The skin reflectance peaks at 800nm, decreases at 900nm, and increases again after reaching the minimum, but the increase is very small. After actual testing, as shown in Figure 1, the reflectivity of the dummy does not have this characteristic. This reflectance property can be used to distinguish real skin from dummies and other items. However, at present, spectrometers are mostly used to obtain reflectance, which is expensive and time-consuming, so it cannot be measured in real time.

[citation list]

[Non-patent literature]

1. Elli Angelopoulou, "Understanding the color of human skin", Computer Science Dept., Stevens Institute of Technology, Castle Pointon Hudson, Hoboken, NJ07030, USA

US6370260 of IoannisPavlidis shows a method of human skin detection using images in two near-infrared bands of 800nm-1400nm and 1400nm-2200nm, using the unique reflection characteristics of human skin in these two bands to distinguish real people from dummy people. However, imaging equipment in these two bands is very expensive and difficult to use in general civilian applications.

Sony's 201080002426 proposes a skin detection system and method using multi-band near-infrared lighting conditions, which can detect multiple pixels of interest in an image and identify corresponding objects that are predetermined. This method uses two different near-infrared wavelengths of 870nm and 950nm to obtain two images at the corresponding wavelengths, and then generates a binarized skin image according to the difference between the two images, and uses the binarized image to extract the skin from the original image. image. This method is mainly used to extract the skin (hand) of the data entry personnel sitting in front of the computer screen during computer data entry, and cannot be applied to skin detection in common scenarios.

Therefore, considering the above situation, it is necessary to find a multi-spectral imaging system with low price and good real-time performance to obtain multi-spectral images. At the same time, there is a need for a skin detection method that can effectively distinguish between real people and fake people in common scenarios.

purpose of invention

The main purpose of the present invention is to provide a multi-spectral imaging system with low price and simple structure, through which the reflection characteristics of objects in the image can be obtained. The present invention provides a detection method for complex backgrounds (especially containing dummy) for skin detection based on the acquired reflection characteristics.

Another object of the present invention is to provide a calibration method, through which the reflection characteristics of objects at various wavelengths are obtained. The function of calibration is to eliminate the difference in spectral response of imaging channels (including light sources, photographing equipment and filters) at different wavelengths, and use a standard reference object with known reflectance at different wavelengths to calibrate the imaging channels at different wavelengths. spectral response to obtain the true reflection properties of the object. The calibration method is related to the reflection characteristics that need to be obtained. Here, the reflection characteristic may refer to the gray value of the pixel, the reflectivity and its deformation, for example, the deformation of the gray value may be the difference between the normalized gray value and the gray value.

The present invention also provides a skin detection method, which regards skin detection as a binary classification problem, uses reflection characteristics as features, and uses machine learning to train a classifier for skin detection.

4. Contents of the invention:

Before the description of the system, method and hardware implementation of the present invention, it should be noted that the present invention is not limited to the described several embodiments and detection methods, and the present invention can have many possible implementations that are not explicitly described herein. example.

The detection method provided by the present invention uses the different reflection characteristics of skin and non-skin under narrow-band multi-spectral irradiation to detect skin, so that human skin and human skin-like objects (including mannequins, masks, photos, etc.) can be effectively distinguished.

The invention provides a multispectral imaging system, which has low cost and good real-time performance. Images under the concerned waveband can be obtained through the system provided by the invention. The light source selected by the system must cover the selected multi-spectral band (natural light can also be used when conditions permit), for example, if the selected bands are 420nm and 850nm, then the selected light source must include these two bands. The shooting device used in this system is a black-and-white camera, and the shooting device can also be selected according to different embodiments, for example, a photodiode can be used as a receiver. In one embodiment of the present invention, a photographing device is used to acquire multispectral images by sequentially switching narrow-band filters of different wavelength bands in front of the photographing device. In addition, two shooting devices can also be used, and narrow-band filters of different bands are installed in front of each shooting device to simultaneously acquire multi-spectral images. It is also possible to use a narrow-band light source or add a narrow-band filter before the light source to obtain multi-spectral images by the shooting device.

The present invention also provides a method for obtaining reflection characteristics in a multi-spectral image. The reflection characteristics include the gray value of the pixel in the band used, the reflectivity and its respective deformation, such as the gray value of the pixel in different bands. Poor, or normalized gray value, reflectance. The present invention mainly uses reflectivity. The present invention obtains approximate reflectance by arranging reference objects in the scene and calibrating. The meaning of calibrating is slightly different according to different embodiments.

An embodiment of obtaining reflection characteristics (reflectivity) in a multispectral image provided by the present invention:

Shooting equipment: black-and-white camera, selected wave bands such as: 420nm, 850nm. Place the narrowband filter in front of the camera and place a reference object of known reflectance in the scene. In order to obtain more image details, the average gray level of the image under each band is between 180-220. The gray level of point P in the image :

${\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; \&Integral;</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; A</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>$

Where i: the i-th band, t: camera exposure time, s: object reflectivity at point P, c: camera response, p: light intensity distribution of multi-spectral light at point P, A: external light at point P The light intensity distribution at , L: filter spectral characteristics. What the present invention uses is narrow-band optical filter, so remove integral sign:

${\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; t</span>}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; A</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>$

Assuming that the light distribution of multi-spectrum and external light is uniform, then the camera response time is a constant value, the camera response c and the filter spectral characteristic L are constant values in the same band, and its I is also the same for the reference object with known reflectance measurable. In this way, it can be obtained

$\mathrm{<span\; class="notranslate">\; tc</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; A</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; ref</span>}}}{{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; ref</span>}}}$

For any pixel in the image, its reflectance:

$\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; ref</span>}}\frac{\mathrm{<span\; class="notranslate">\; I</span>}}{{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; ref</span>}}}$

Under ideal conditions, such as uniform illumination and sufficiently narrow bandwidth of the narrow-band filter, the reflectance obtained by this embodiment is approximately equal to the true reflectance. However, the obtained reflectance provided by this method is mainly used for the skin detection in the present invention, so its accuracy is acceptable. In skin detection, the obtained reflectance or its deformation, such as normalization, will be used as features for training.

Another embodiment provided by the present invention for obtaining reflectance characteristics (reflectivity trend) in multispectral images:

Shooting equipment: black and white video camera, selected wave band: 420nm, 850nm. Place the narrowband filter in front of the multispectral light source, and place a reference object of known reflectance in the scene for the calibration process before shooting. The gray level of point P in the captured image :

${\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; \&Integral;</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; A</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>$

Without taking into account the external light

${\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; idark</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; \&Integral;</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>$

Calibration: From the above formula, t is a fixed value. For a reference object with known reflectivity, adjust the value of p ^P (λ)L(λ) by adjusting the light intensity of the light source in the selected band, and ensure that the average image in the selected band Under the condition that the gray value is 180-220, the values of tc(λ)p ^P (λ)L(λ) in each band are equal. In this way, under the condition of no external light, the trend of pixel gray value can represent the trend of reflectance, and even the gray value multiplied by a weight factor w can represent the reflectance in the corresponding band. However, this method does not work under the condition of external light. In order to solve this problem, this embodiment removes the influence of external light by designing the deformation of the reflection characteristics.

After the calibration process is completed, in the case of external light, the influence of external light can be eliminated by the following formula.

${\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; \&Integral;</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; A</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; \&Integral;</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; A</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>$ $<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; \&Integral;</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; P</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>$

The difference between pixels in different bands is used as the characteristic of the pixel, and this characteristic is used as the training feature. In order to further reduce the influence of external light, the difference can be normalized.

Through the method provided by the invention, the reflection characteristics of the pixels in the image can be obtained according to the multispectral image obtained by the system. Depending on the embodiment, the characteristics of the pixels to be calculated are different.

The present invention also provides a method for skin detection, which extracts the reflection characteristics of positive samples (skin) and negative samples (non-skin objects) in multispectral images as features, and obtains training sets and test sets. Use machine learning methods (such as SVM, ADABOOST) to train and obtain a classifier, and then use the classifier for skin detection.

Description of drawings:

Figure 1 Comparison of reflectance between skin and dummy

Figure 2 System structure

Figure 3 training process

Figure 4 Captured images, examples of positive and negative samples at 420nm and 800nm

Figure 5 The actual reflectance obtained

Figure 6 detection process

Figure 7 test results

Figure 8 Reference object reflectance

detailed description:

Device connection:

The processor 300 is connected to the light source 100 through the controller 101 to control the switch of the light source and output brightness, and the processor 300 is connected to the shooting device through the capture card. Since the purpose of shooting is to acquire multi-spectral images, the processor 300 should be able to ensure that the shooting device can capture images synchronously when the light sources of each wavelength band are turned on. For example, if you want to obtain images in two bands of 420nm and 850nm. The processor 300 should be able to ensure that when the light source of the 420nm band is turned on, the shooting device starts shooting, and when the light source of the 420nm band is turned off, the shooting device stops shooting.

Scene layout:

According to the method provided by the present invention, in order to obtain the reflection characteristics of the pixels in the image, it is necessary to place a reference object with known reflectance in the scene to obtain the reflection characteristics of the pixels in the image. The present invention uses a PTFE plate as a reference.

Since the present invention provides various embodiments, and the specific implementation manners of different embodiments are slightly different, only the implementation manners of the above-mentioned two embodiments are described here. In the experiments carried out in the present invention, the selected light source is a xenon lamp with a spectral distribution of 350nm-2500nm (natural light can be selected depending on the embodiment), the selected wave band is a narrow-band filter of 420nm and 850nm, the bandwidth is 20nm, and the shooting equipment is black and white camera.

Embodiment one:

After the equipment is connected and the scene layout is completed, put the filter 401 (420nm) in front of the camera 200 . Then perform the calibration process: adjust the light intensity of the light source 100 through the processor 300 so that the gray value of the scene to be photographed in the picture is between 180-220. The process of taking an 850nm image (denoted as pic2) is the same as the above process, only need to replace the filter.

Sample extraction: Select skin and non-skin pixels as positive and negative samples respectively.

Feature extraction: According to the method provided in [0016], calculate the reflectivity characteristics of the pixel under the selected waveband, then the pixel feature value is a two-dimensional vector composed of reflectivity under the waveband 420nm and waveband 800nm.

Training: Calculate and mark all the eigenvalues of the selected samples, and use SVM training to obtain a classifier.

Detection: Obtain pic1 and pic2 in the same way, calculate the feature value of each pixel, and use the classifier to detect the final result

Embodiment two:

After the equipment is connected and the scene layout is completed, put the filter 401 (420nm) in front of the light source 100 . Then perform calibration, the calibration principle and method are as shown in [0021], after the calibration is completed, the image taken under the wavelength band 420nm is recorded as pic1, it should be noted that the calibration of this embodiment needs to be carried out under the condition of darkness (no external light) . Get pic2 in the same way.

Sample extraction: Select skin and non-skin pixels as positive and negative samples respectively.

Feature extraction: the purpose of the calibration in this embodiment is to use the gray value trend of a pixel to characterize its reflectance trend, and then use the normalized gray value difference as the feature of the pixel. That is, the feature of this embodiment is one-dimensional.

Training: Calculate and mark all the eigenvalues of the selected samples, and use SVM training to obtain a classifier.

Detection: Obtain pic1 and pic2 in the same way, calculate the feature value of each pixel, and use the classifier to detect to obtain the final result.

Claims (5)

1., by obtaining the method that in two wave band multispectral images, object reflectance carries out human body skin detection in multi-optical spectrum imaging system, said method comprising the steps of:

1a) shooting: be used for shooting multispectral image before the optical filter of selected two wave bands is put in capture apparatus camera lens respectively, and in the scene that is taken, place object of reference, for obtaining the reflection characteristic of each pixel in image, make the average gray of image to ensure not lose image information as far as possible between 180-220 by regulating multispectral light source controller change light-source brightness；

1b) calibration: utilize object of reference reflection characteristic obtain other object reflectance characteristics in image to implement step as follows:

1. processor obtains the gray value of certain pixel of object of reference in image or some region of average gray, and reads this reflection characteristic of object of reference reflectance preserved；

2. processor reads the grey scale pixel value of shooting objects in images；

3. processor calculates the reflectivity Characteristics obtaining objects in images according to below equation:

1c) training: extract the positive sample of skin and non-skin object negative sample from the picture of shooting, input as grader using the reflectivity Characteristics of each pixel, utilizes the method training grader of machine learning；

1d) detection: the grader obtained with the training stage carries out human body skin detection.

2., by obtaining the method that in two wave band multispectral images, object difference reflectance signature carries out human body skin detection in multi-optical spectrum imaging system, said method comprising the steps of:

2a) shooting: to obtain narrow-band light source before the optical filter of selected two wave bands is put in multispectral light source respectively, and in the scene that is taken, place object of reference, for obtaining the reflection characteristic of each pixel in image；

2b) calibration: utilize the step that realizes that object of reference reflection characteristic obtains objects in images difference reflectance signature to include:

1. under the dark surrounds without ambient light, processor obtains certain grey scale pixel value of object of reference position in the spectrum picture of first wave band or a certain area grayscale meansigma methods by capture apparatus, read this reflection characteristic of object of reference reflectance preserved, and the ratio both calculating, preserve result of calculation；

2. the intensity of illumination that processor changes under second band by regulating multispectral light source controller makes gray value under this wave band equal with the result under the ratio of reflectance and first wave band, it may be assumed that

3. after completing Illumination adjusting under dark surrounds, in the shooting environmental having ambient light, same pixel difference under two wave bands is utilized to make a return journey except the impact of ambient light, it is thus achieved that object difference reflectance signature, in order to difference is normalized by the impact weakening ambient light further uneven brought；

2c) training: extract the positive sample of skin and non-skin object negative sample from the picture of shooting, input as grader using the object difference reflectance signature of each pixel, utilizes the method training of machine learning to obtain grader；

2d) detection: the grader obtained with the training stage carries out human body skin detection.

3. method as described in claim 1 or 2, needs in the scene that is taken to place object of reference, and the reflection characteristic of object of reference must be known.

4. method as described in claim 1 or 2, multispectral image under acquired two wave bands, a wave band is at 420nm, and another wave band is between 780-850nm.

5. method as described in claim 1 or 2, considers skin detection as two classification problems, calculates the reflection characteristic of each pixel in multispectral image when detection, then obtains final result with grader classification.