Traditional Chinese medicine face image color quantitative characterization method fusing color space and texture features

Abstract

The invention discloses a traditional Chinese medicine face image color quantitative characterization method fusing color space and texture features, and relates to the technical field of image processing and medical image analysis. The method comprises the following steps: performing illumination normalization and noise suppression processing on a facial image of a patient to obtain an enhanced and denoised image, and constructing a structure response graph and a color fusion feature tensor; introducing a structure-guided color enhancement mechanism, and obtaining a structure-enhanced multi-channel color tensor based on the structure response diagram and the color fusion feature tensor; constructing a yellow saliency guide map and a multi-scale texture response tensor, and generating a joint feature map through a color-texture joint decoupling mechanism under saliency weighting; normalizing the combined feature map, and extracting texture features and colors of the normalized combined feature map; and combining the color and texture features into a joint quantitative feature vector, and quantitatively representing the color of the traditional Chinese medicine facial image. The problems that color distinguishing is unstable and color and texture decoupling is difficult under complex illumination are solved.

Description

Technical Field

[0001] This invention relates to the fields of image processing and medical image analysis technology, and in particular to a method for quantitative characterization of color in traditional Chinese medicine facial images by integrating color space and texture features. Background Technology

[0002] Traditional Chinese medicine (TCM) facial diagnosis, the first of the four diagnostic methods (inspection, auscultation, inquiry, and palpation), is a traditional method of judging a person's health status by observing facial features such as color, shape, and texture. Among these, accurate identification of facial color is the core of facial diagnosis. TCM theory clearly divides facial color into "normal color," which is the normal color in a healthy state, such as a bright, moist, and subtle yellow hue, and "pathological color," which is an abnormal color in a pathological state, such as a dull, lackluster, and withered yellow hue. The subtle differences between these two types of color directly affect the accuracy of the diagnostic conclusion. Currently, image-based quantitative research on TCM facial diagnosis has become a hot topic, but existing quantitative methods still have significant shortcomings: (1) Poor color space adaptability: Under complex natural lighting conditions, the conventional RGB color space is easily affected by the intensity of light. Although the HSV color space can separate hue, saturation, and brightness, it is not stable enough in distinguishing between "pathological yellow" and "physiological yellow" in human facial skin, making it difficult to capture the subtle hue differences between the two; (2) Difficulty in decoupling color and texture: Skin gloss, as a key texture feature, has a strong coupling relationship with color information. Existing technologies mostly use a single color channel or simple texture statistics for characterization, which cannot achieve effective decoupling of texture and color information, resulting in the quantitative accuracy of pathological color being severely affected by gloss. Therefore, in order to address the above shortcomings, it is urgent to provide a facial image color quantification method that integrates color space and texture features. Summary of the Invention

[0003] This invention provides a method for quantitative characterization of color in TCM facial images that integrates color space and texture features, in order to solve the problems of unstable color differentiation and difficulty in decoupling color and texture under complex lighting conditions.

[0004] The present invention provides a method for quantitative characterization of color in traditional Chinese medicine facial images by integrating color space and texture features, comprising the following steps: S1. Acquire patient facial images, perform illumination normalization and noise suppression processing to obtain enhanced and denoised images; based on the enhanced and denoised images, construct a structure response map and a color fusion feature tensor; introduce a structure-guided color enhancement mechanism, and based on the structure response map and the color fusion feature tensor, obtain a structure-enhanced multi-channel color tensor; S2. Construct a yellow saliency guide map and a multi-scale texture response tensor. Generate a joint feature map through a color-texture joint decoupling mechanism under saliency weighting. Normalize the joint feature map and extract the texture features and color of the normalized joint feature map. Combine the color and texture features into a joint quantized feature vector to quantify the color of TCM facial images.

[0005] Preferably, S1 specifically includes: The enhanced and denoised images were then converted to CIE L. * a * b * Color space and YCbCr color space.

[0006] Preferably, S1 specifically includes: From CIE L * a * b * Extracting b from color space * Channel pixel values, and C values ​​extracted from the YCbCr color space. b Channel pixel values; concatenate the two channels along the channel dimension to generate a color fusion feature tensor.

[0007] Preferably, S1 specifically includes: In the implementation of the structure-guided color enhancement mechanism, the structure response map is used as the guiding basis for CIEL. * a * b * b in color space * C in the channel and YCbCr color space b The channels are weighted and enhanced using channel enhancement factors to obtain a structurally enhanced multichannel color tensor.

[0008] Preferably, S2 specifically includes: Introducing a yellow saliency guide map, the weights are fused with the normalized b through color channels. * Channel pixel value, C b The weighted combination of channel pixel values ​​generates a patient's facial image that exhibits a significant response intensity of pathological yellow tones.

[0009] Preferably, S2 specifically includes: Based on the structurally enhanced multi-channel color tensor, a single-channel fused image is constructed; based on the single-channel fused image, a multi-scale image is generated, and directional texture responses are extracted from the images at each scale to generate a multi-scale texture response tensor.

[0010] Preferably, S2 specifically includes: In the implementation of the color-texture joint decoupling mechanism under saliency weighting, a color feature map is constructed, and the multi-scale texture response tensor is averaged in the channel dimension. Then, pixel-level weighted fusion is performed through the yellow saliency guide map to generate a joint feature map.

[0011] Preferably, S2 specifically includes: The joint feature map is normalized, and the normalized joint feature map is replaced with a grayscale image. The grayscale image is divided into two or more regions, and the texture statistics of texture features are calculated in each region. At the same time, the pixel mean of each region is extracted as the main yellow color. Finally, the color and texture statistics are combined into a joint quantized feature vector for each region.

[0012] The beneficial effects of the technical solution of the present invention are: 1. This invention addresses the problem of traditional RGB color models' difficulty in distinguishing between "pathological yellow" and "physiological yellow" by introducing CIE L (Cipher Illustrated Guided Color). * a * b * A structure-guided color enhancement mechanism is proposed, which integrates the YCbCr dual color space fusion mechanism and controls the color enhancement operation through the structural information of the patient's facial image itself. This enables pathological tones to have a stronger response in structurally significant areas and remain suppressed in non-diagnostic areas, thereby achieving synergistic enhancement of the patient's facial structure and color and effectively improving the separation and positioning accuracy of pathological tones.

[0013] 2. This invention introduces a color-texture joint decoupling mechanism under saliency weighting, using the yellow saliency guide map as a weight to synergistically fuse the color feature map and the average texture response intensity, thereby achieving enhanced response and dynamic decoupling of color-texture joint perception in the image. This makes pathological color regions not only highly expressive in terms of color features, but also have accurate modeling capabilities in terms of texture features, effectively improving the completeness of the expression of complex color and structural composite features in TCM facial images. Attached Figure Description

[0014] Figure 1 This is a flowchart of a method for quantitative characterization of color in traditional Chinese medicine facial images that integrates color space and texture features, as described in this invention. Detailed Implementation

[0015] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0017] The following description, in conjunction with the accompanying drawings, details a specific scheme for a method of quantitative characterization of color in traditional Chinese medicine facial images that integrates color space and texture features, provided by the present invention.

[0018] See attached document Figure 1 The diagram illustrates a flowchart of a method for quantifying the color of facial images in Traditional Chinese Medicine by fusing color space and texture features, according to an embodiment of the present invention. The method includes the following steps: S1. Acquire patient facial images, perform illumination normalization and noise suppression processing to obtain enhanced and denoised images; based on the enhanced and denoised images, construct a structural response map and a color fusion feature tensor; introduce a structure-guided color enhancement mechanism, and based on the structural response map and the color fusion feature tensor, obtain a structure-enhanced multi-channel color tensor.

[0019] Patient facial images in RGB color space were acquired and subjected to illumination normalization and noise suppression. Illumination normalization employed existing image enhancement techniques, such as the multi-scale Retinex algorithm, to enhance color consistency of the patient's facial images under complex acquisition conditions including shadows, sidelighting, and reflections, resulting in an enhanced image. To further remove high-frequency noise introduced by the acquisition device or environment from the enhanced image, a bilateral filtering method was used for smoothing, yielding an enhanced and denoised image. .

[0020] Based on the enhanced and denoised image The classical Sobel operator is used to obtain a structural response map, which is used to numerically reflect the structural complexity of the patient's facial image. The classical Sobel operator uses a horizontal convolution kernel. Convolution kernel in the vertical direction Calculate the structure response map at the pixel location Response value at , Indicates pixel position. A response value closer to 1 indicates a higher pixel position. Located in structurally significant areas such as skin edges, patches, wrinkles, or pigmentation transition zones, a response value closer to 0 indicates a more prominent pixel location. Located in a relatively smooth area with weak texture or in the background area.

[0021] Furthermore, the denoised image will be enhanced. Using existing nonlinear and linear color space conversion methods, the color space can be converted to CIE L. * a * b * Color space and YCbCr color space. Nonlinear color space conversion methods enhance and denoise the image using a standard linear transformation matrix. The RGB values ​​are converted to the CIE XYZ color space, and then non-linear functions, such as the cube root function and piecewise linear functions, are introduced to convert the CIE XYZ color space to the CIE L color space. * a * b * Non-linear color space conversion. Linear color space conversion methods enhance the denoised image through linear weighting and channel offset. A linear transformation from RGB values ​​to the YCbCr color space. In the application scenario of TCM facial diagnosis images, the red difference component (Cr) is easily affected by individual differences in blood color and physiological state, while the blue difference component (Cb) shows more stable statistical characteristics in describing the yellow-blue trend in skin color. Therefore, Cb is selected. b The channel serves as a representative channel in the YCbCr space for modeling the saliency of pathological yellowing. (From CIE L) * a * b * Extracting b from color space * Channel pixel value This reflects the difference in blue and yellow tones; negative pixel values ​​represent blue, and positive pixel values ​​represent yellow, with larger pixel values ​​indicating a more yellow tint. Simultaneously, C is extracted from the YCbCr color space. b Channel pixel value It is highly sensitive to the bluish tendency of yellow tones in skin tones, and the smaller the pixel value, the more yellow it appears; the two channels mentioned above are concatenated along the channel dimension to construct a color fusion feature tensor. .

[0022] To enhance the denoised image To enhance the high response of the yellow region in pathological studies while avoiding false enhancement of irrelevant regions, a structure-guided color enhancement mechanism is proposed: using the structure response map as a guide, the two color channels representing the yellow hue—CIEL—are enhanced. * a * b * b in color space *C in the channel and YCbCr color space b The channels are weighted and enhanced using a channel enhancement factor to obtain the structure-enhanced multi-channel color tensor, as shown in the following formula: , in, This represents the enhanced multi-channel color tensor at the pixel location. and channels Pixel value at; , representing the channel index, respectively representing b * Channel, C b aisle; This represents the channel enhancement factor, which is the response value of the structure-response plot. When it is 0, that is, the pixel position No color enhancement is performed when the area is in a smooth region; The structural enhancement coefficient is used to control the influence of the structural response map on color enhancement. It is obtained by statistically analyzing the average value of the structural response map. Represents the structural response map at pixel location The response value at the location; Indicates pixel position b at the location * Channel pixel values; Indicates pixel position C at the location b Channel pixel value.

[0023] The above formula proposes a structure-guided color enhancement mechanism that controls the color enhancement operation by using the structural information of the patient's facial image itself. This allows pathological tones to have a stronger response in structurally significant areas and remain suppressed in non-diagnostic areas, achieving synergistic enhancement of the patient's facial structure and color, and effectively improving the separation and positioning accuracy of pathological tones.

[0024] S2. Construct a yellow saliency guide map and a multi-scale texture response tensor. Generate a joint feature map through a color-texture joint decoupling mechanism under saliency weighting. Normalize the joint feature map and extract the texture features and color of the normalized joint feature map. Combine the color and texture features into a joint quantized feature vector to quantify the color of TCM facial images.

[0025] To focus texture modeling on diagnostically valuable yellow regions, a yellow saliency guide map is introduced, which integrates weights from color channels with normalized pixel positions. b at the location * Channel pixel value, C b Weighted combination of channel pixel values ​​to generate a patient's facial image at pixel location The formula for the significant response intensity of the pathological yellow hue at the location is as follows: , in, The yellow salience guide image is located at the pixel position. The normalized pixel value at that location represents the pixel position of the patient's facial image. The area exhibits a significant response intensity of pathological yellow hue. The closer to 1, the higher the pixel position. The color characteristics are closer to pathological yellow, which is more worthy of attention in texture analysis; The color channel blending weights are represented by b. * Channel, C b The contribution ratio of each channel in constructing the yellow saliency guide map satisfies This was achieved through cross-validation and comparison of historical patient facial image data, sourced from existing medical databases such as the TCM-FD (Traditional Chinese Medicine Facial Diagnosis) Dataset. Specifically, based on the TCM-FD Dataset, 500 patient facial images tagged with "pathological yellow" were selected. Traditional Chinese medicine experts manually and precisely annotated the pathological color areas in the images, generating a mask of the true pathological yellow areas. (The last sentence appears to be incomplete and possibly refers to further details about the process.) * Channel and C b After channel enhancement, denoising, and structure-guided color enhancement, the respective yellow response regions are extracted and compared with the actual pathological yellow region mask labeled by traditional Chinese medicine experts using IoU (Intersection over Union). The results yield b... * Channel and C b The IoU value corresponding to the channel, and respectively b * Channel, C b The IoU value corresponding to the channel is in b * Channel and C b The proportion of the sum of IoU values ​​corresponding to each channel is used as b. * Channel and C b The passage is Color channel blending weights during the construction process; Represents the normalized pixel position b at the location * Channel pixel value, used to measure b * The saliency response intensity of the pathological yellow hue of the channel was determined by the structure-enhanced multichannel color tensor. Pixel value of the first channel Obtained using the Min-Max normalization method; Indicates the normalized pixel position C at the location bThe saliency of the channel pixel value converted to the pathological yellow tint is related to b. * The yellow trend represented by the channel remains consistent in direction; Represents the normalized pixel position C at the location b Channel pixel value, used to measure C b The degree of blue tint in a channel; a smaller value indicates less blue content and a more yellowish tint. This is determined by the multi-channel color tensor after structural enhancement. Pixel value of the second channel It was obtained using the Min-Max normalization method.

[0026] To effectively capture the skin texture variation characteristics at different spatial scales and orientations, middle and A single-channel fused image is constructed through linear weighted fusion, where the weights are obtained by evaluating the intersection-union ratio (IU) of historical patient facial image data. Based on the single-channel fused image, a multi-scale image is generated using the existing difference-of-Gaussian pyramid method. A standard Gabor filter is then used to extract directional texture responses at each scale, and finally, the images are stitched together according to the texture response dimension to form a unified multi-scale texture response tensor. ,in Indicates pixel position, Indicates the texture response dimension.

[0027] To achieve explicit decoupling between color and texture, a saliency-weighted joint decoupling mechanism for color and texture is proposed: constructing a color feature map. Simultaneously, for multi-scale texture response tensors The average is applied across the channel dimension, and pixel-level weighted fusion is performed using a yellow saliency guide map, as shown in the following formula: , in, This is a joint feature map, representing the pixel location. The perceived intensity of color and texture is integrated at the point of contact; Indicates the pixel position The intensity of perception dominated by color; The color feature control weights were obtained through five-fold cross-validation on historical patient facial image data. A higher value indicates stronger color dominance. Specifically, facial images of 500 patients with "pathological yellow" markings from the TCM-FD Dataset were randomly divided into five subsets to ensure balanced sample distribution. In each round of cross-validation, four subsets were selected as the training set, and the remaining subset as the validation set. This process was repeated five times to ensure each image served as a validation sample. For each training fold: a color binary classification sub-model with only color features and a texture binary classification sub-model with only texture features were constructed. The recognition performance of the two sub-models on the pathological yellow region in the validation set was compared, and the F1-score was calculated for each. After five-fold training, the average F1-score of the color and texture binary classification sub-models was calculated. The color feature control weights were obtained by calculating the proportion of the average F1-score of the color binary classification sub-model in the sum of the average F1-scores of the color and texture binary classification sub-models. The yellow salience guide image is located at the pixel position. The normalized pixel value at that point is used as a pixel-level attention weight to control the participation of color and texture features; This is a color feature map used to describe the current pixel position. The yellow hue intensity is taken from the structure-enhanced multichannel color tensor. The first channel reflects the difference in blue and yellow tones; Indicates the pixel position Perceived intensity dominated by texture; For texture response dimensions; The weights are controlled for texture features; the larger the value, the stronger the dominance of texture. For multi-scale texture response tensors; Indicates pixel position The average texture response intensity.

[0028] The above formula, through a color-texture joint decoupling mechanism under saliency weighting, enables pathological color regions to not only have high expressiveness in color features, but also accurate modeling ability in texture features, effectively improving the expressive integrity of complex color and structural composite features in TCM facial images.

[0029] Structured representation of texture patterns in the joint feature map: Min-Max normalization is applied to the joint feature map. Normalization is performed, and the normalized joint feature map is converted into a grayscale image using existing grayscale quantization methods. The grayscale image is then divided into sections using a sliding window approach. Each region, in the region The region is calculated using the existing gray-level co-occurrence matrix method. Contrast of texture features ,energy With entropy Texture statistics; simultaneously, extracting regions Corresponding pixel position average pixel value As a region Yellow as the primary color; the average pixel value By analyzing the region b * The sum of pixel values ​​at all pixel locations in the channel, i.e. Then with the region The result is obtained by dividing the number of pixel locations contained within; finally, the color and texture statistics are combined into a region. Joint quantized eigenvectors As a quantitative representation of the color of facial images in traditional Chinese medicine that integrates color space and texture features.

[0030] In summary, a method for quantitative characterization of color in TCM facial images that integrates color space and texture features has been developed.

[0031] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0032] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0033] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for quantitatively representing the color of facial images in Traditional Chinese Medicine by integrating color space and texture features, characterized in that, Includes the following steps: S1. Acquire patient facial images, perform illumination normalization and noise suppression processing to obtain enhanced and denoised images; Based on the enhanced and denoised image, construct the structural response map and the color fusion feature tensor; A structure-guided color enhancement mechanism is introduced, and a structure-enhanced multi-channel color tensor is obtained based on the structure response map and the color fusion feature tensor. S2. Construct a yellow saliency guide map and a multi-scale texture response tensor. Generate a joint feature map through a color-texture joint decoupling mechanism under saliency weighting. Normalize the joint feature map and extract the texture features and color of the normalized joint feature map. Combine the color and texture features into a joint quantized feature vector to quantify the color of TCM facial images.

2. The method for quantitative characterization of color in TCM facial images by integrating color space and texture features according to claim 1, characterized in that, S1 specifically includes: The enhanced and denoised images were then converted to CIE L. * a * b * Color space and YCbCr color space.

3. The method for quantitative characterization of color in TCM facial images by integrating color space and texture features according to claim 2, characterized in that, S1 specifically includes: From CIE L * a * b * Extracting b from color space * Channel pixel values, and C values ​​extracted from the YCbCr color space. b Channel pixel values; concatenate the two channels along the channel dimension to generate a color fusion feature tensor.

4. The method for quantitative characterization of color in TCM facial images by integrating color space and texture features according to claim 3, characterized in that, S1 specifically includes: In the implementation of the structure-guided color enhancement mechanism, the structure response map is used as the guiding basis for CIE L. * a * b * b in color space * C in the channel and YCbCr color space b The channels are weighted and enhanced using channel enhancement factors to obtain a structurally enhanced multichannel color tensor.

5. The method for quantitative characterization of color in TCM facial images by integrating color space and texture features according to claim 1, characterized in that, S2 specifically includes: Introducing a yellow saliency guide map, the weights are fused with the normalized b through color channels. * Channel pixel value, C b The weighted combination of channel pixel values ​​generates a patient's facial image that exhibits a significant response intensity of pathological yellow tones.

6. The method for quantitative characterization of color in TCM facial images by fusing color space and texture features according to claim 1, characterized in that, S2 specifically includes: Based on the structurally enhanced multi-channel color tensor, a single-channel fused image is constructed; based on the single-channel fused image, a multi-scale image is generated, and directional texture responses are extracted from the images at each scale to generate a multi-scale texture response tensor.

7. The method for quantitative characterization of color in TCM facial images by integrating color space and texture features according to claim 6, characterized in that, S2 specifically includes: In the implementation of the color-texture joint decoupling mechanism under saliency weighting, a color feature map is constructed, and the multi-scale texture response tensor is averaged in the channel dimension. Then, pixel-level weighted fusion is performed through the yellow saliency guide map to generate a joint feature map.

8. The method for quantitative characterization of color in TCM facial images by integrating color space and texture features according to claim 7, characterized in that, S2 specifically includes: The joint feature map is normalized, and the normalized joint feature map is replaced with a grayscale image. The grayscale image is divided into two or more regions, and the texture statistics of texture features are calculated in each region. At the same time, the pixel mean of each region is extracted as the main yellow color. Finally, the color and texture statistics are combined into a joint quantized feature vector for each region.

Images