CN117197064A - Automatic non-contact eye red degree analysis method - Google Patents

Abstract

The invention discloses a non-contact automatic analysis method of eye redness, which includes the following steps: aiming at the front of the face to capture human eyes, performing human eye area detection and image preprocessing on the captured images, scaling and preprocessing, and then using the U‑Net++ model. Perform scleral area detection to obtain a binary image, and operate the binary image and the zoom image to obtain a color image of the scleral area; perform image smoothing and adaptive histogram equalization to enhance image contrast, and use the B‑COSFIRE filter to The image enhancement map is filtered, and at the same time, the image enhancement map is converted into a LAB color model to obtain the image mask. Finally, the filtered map and the image mask are ANDed to obtain the bloodshot binary image; the scleral area color map and The number of non-zero pixels in the bloodshot binary image is then calculated as a ratio to obtain the ratio of jealousy. The invention provides a non-contact automatic analysis method for the degree of eye redness, which has high adaptability and accuracy in automatic determination and can realize intelligent auxiliary diagnosis of the patient's degree of eye redness.

Description

A non-contact automatic analysis method of eye redness degree

Technical field

The invention relates to a non-contact automatic analysis method of eye redness, and belongs to the technical fields of image processing and medical auxiliary diagnosis.

Background technique

Eye redness analysis can determine the severity of ocular surface inflammation. As one of the important indicators for dry eye detection, it plays an important role in the field of ophthalmic medical diagnosis. Traditional eye redness analysis is judged by doctors based on experience, which has the disadvantage of large errors. In addition, although digital image processing methods have emerged, the current digital image processing methods have low adaptability and accuracy in automatic determination, and cannot achieve intelligent auxiliary diagnosis of the patient's degree of redness.

Contents of the invention

The technical problem to be solved by the present invention is to overcome the shortcomings of the existing technology and provide a non-contact automatic analysis method for the degree of eye redness, which has high adaptability and accuracy in automatic determination and can realize intelligent auxiliary diagnosis of the patient's degree of eye redness.

In order to solve the above technical problems, the technical solutions adopted by the present invention are:

A non-contact automated eye redness analysis method, including the following steps:

Use visible light imaging equipment to shoot the human eyes from the front of the face, perform human eye area detection and image preprocessing on the captured images to obtain preprocessed images;

The preprocessed image is scaled to obtain a zoomed image. After preprocessing the zoomed image, the U-Net++ model is used to detect the scleral area, and a binary image of the scleral area of the image to be inspected is obtained. The binary image is combined with the zoomed image. Get a color map of the sclera area;

Perform image smoothing and adaptive histogram equalization on the scleral area color map to enhance image contrast and obtain an image enhancement map. Then use the B-COSFIRE filter to filter the image enhancement map to obtain a filtered map. At the same time, the image is enhanced. The image is converted into a LAB color model, and the image mask is obtained. Finally, the filtered image and the image mask are ANDed to obtain the bloodshot binary image;

Calculate the number of non-zero pixels in the color map of the sclera area and the binary blood map respectively, and then calculate the ratio of the number of non-zero pixels in the binary blood map and the color map of the sclera area to obtain the proportion of eye redness. .

The human eye area detection includes the following steps:

When detecting the captured image, use the trained classifier to search area by area starting from the upper left corner of the captured image, and use the similarity criterion to determine whether it is a human eye;

If the judgment result is human eyes, select the range where the eye area is located and determine whether the eye area range is no less than 256×256. If so, proceed to the next step. If not, it will prompt that the human eye image is too small and require a new process. Photography;

If the judgment result is not human eyes, it will prompt that the input image is wrong and require re-shooting;

Among them, the training process of the human eye classifier is: using the deep learning method, first collect a large number of human eye and non-human eye samples from the Internet for pre-training, and obtain the deep learning model parameters to build the human eye classifier.

Taking images for preprocessing involves the following steps:

According to the result of the eye area frame selection, the length and width of the human eye area are expanded by 1.2 times according to the original proportion, so that the final eye area selected by the frame contains complete eye information at the corner of the eye. The expanded length is used as the benchmark, and the parallel center line is the baseline. , cropped with an aspect ratio of 5:4 so that all eye images meet the size requirements and maximize the effective area.

The zoom image preprocessing includes the following steps: using the color threshold method, if the overall color of the image is judged to be reddish, extract the R channel of the three RGB channels of the image and continue the scleral segmentation operation; otherwise, use the histogram equalization method for image enhancement Finally, the scleral segmentation operation is still performed using the RGB three-channel color image.

The U-Net++ model is an improved U-Net++ model. The input image is a zoomed image, the output image is a single-channel binary image, and the input and output image shapes are both 512×512; the convolution operation of each layer of U-Net++ is the same as Between the ReLU activation function operations, a normalization operation is added; between the upper layer and the lower layer, a dropout operation is added, and an attention mechanism is added afterwards, followed by a pooling operation.

The specific operation of the binary image and the zoom image is as follows: set all the pixels with a pixel value of 0 in the binary image to 0 for the corresponding pixel values in the zoom image, and only retain the sclera area in the image to obtain the sclera area map.

The specific calculation method of jealousy proportion is as follows:

Among them, Degree represents the proportion of jealousy, B _pixels represents the number of non-0 pixels in the bloodshot binary image, and S _pixels represents the number of non-0 pixels in the sclera area map.

Multiply the calculation result of the jealousy proportion by 100 and round down to the nearest whole, so that the final result is within the range of [0,100]. Finally, use the fractional scale to obtain the corresponding jealousy level.

A non-contact automatic eye redness analysis device, including:

The eye image acquisition and processing module is used to use visible light imaging equipment to shoot human eyes from the front of the face, and perform human eye area detection and image preprocessing on the captured images to obtain preprocessed images;

The scleral area detection module is used to scale the preprocessed image to obtain a zoomed image. After preprocessing the zoomed image, use the U-Net++ model to detect the scleral area of the zoomed image to obtain a binary image of the scleral area of the image to be inspected. Perform operations on the binary image and the zoom image to obtain a color image of the sclera area;

The eye-red area extraction module is used to extract the color image of the sclera area, perform image smoothing and adaptive histogram equalization to enhance the image contrast, and obtain the image enhancement map, and then use the B-COSFIRE filter to filter the image enhancement map to obtain Filter the processing map, and at the same time convert the image enhancement map into a LAB color model to obtain the image mask. Finally, perform the AND operation on the filtering processing map and the image mask to obtain the bloodshot binary image;

The eye red proportion calculation module is used to calculate the number of non-zero pixels in the sclera area color image and the bloodshot binary image respectively, and then calculate the number of non-zero pixels in the bloodshot binary image and the sclera area color image. Perform a ratio calculation to get the jealousy ratio.

A computer-readable storage medium has a computer program stored thereon. When the computer program is executed by a processor, the method for automatic analysis of the degree of contact envy is implemented.

Beneficial effects of the present invention: The present invention provides a non-contact automatic analysis method of eye redness, using visible light imaging equipment to aim at the front of the face to photograph human eyes, and using the camera function of an ordinary smartphone or camera to achieve non-contact eye images Acquisition, no auxiliary light source, simple structure, no discomfort; it can avoid errors caused by eye redness appearing areas and congestion around the eyes, accurate eye red position and area, provide accurate data, and automatically determine the adaptability and accuracy.

Description of the drawings

Figure 1 is a work flow chart of a non-contact automatic eye redness analysis method according to the present invention.

Figure 2 is a flow chart of eye image acquisition and processing in the present invention;

Figure 3 is a flow chart of sclera area detection in the present invention;

Figure 4 is a flow chart of eye red area extraction in the present invention;

Figure 5 is a flow chart for calculating the ratio of eye redness and judging the degree of eye redness in the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings. The following examples are only used to illustrate the technical solutions of the present invention more clearly, but cannot be used to limit the scope of protection of the present invention. As shown in Figure 1, the present invention discloses a non-contact automatic analysis method for eye redness, which includes the following steps:

Step 1: Eye image acquisition and processing.

Use a visible light imaging device to shoot the human eye from the front of the face, and the image must contain a complete single human eye. Due to different resolutions of imaging equipment, it will affect the video memory occupied by scleral area detection neural network training, scleral area detection results and algorithm execution efficiency. Human eye area detection and image preprocessing must be performed first. Using visible light imaging equipment to aim at the front of the face to take pictures of human eyes can be achieved by using the camera function of an ordinary smartphone or camera. It only needs to satisfy that the image size of the eye area is not less than 256×256, and does not require additional equipment or special ophthalmology equipment. instrument.

For human eye area detection, by pre-training a large number of collected human eye and non-human eye samples, the model parameters are obtained to build a human eye classifier. When detecting the captured image, use the trained classifier to search region by region starting from the upper left corner of the captured image, and use the similarity criterion to determine whether it is the human eye. Area-by-area search is to divide the input image into several small rectangular areas, and perform human eye detection and judgment on each small area. The similarity criterion refers to using the human eye classifier model to calculate the similarity features of the objects detected in each small area during the area-by-area search process. The calculation results represent the degree of similarity between the detected objects and the human eye.

If the judgment result is human eyes, you also need to frame the range of the eye area. The range of the eye area is not less than 256×256. Then continue to the next step of image preprocessing. Otherwise, it will prompt that the human eye image is too small and require re-shooting. If the judgment result is not human eyes, it will prompt that the input image is wrong and require re-shooting. Image preprocessing is to expand the length and width of the human eye area by 1.2 times according to the original proportion based on the eye area frame selection results, so that the final frame-selected eye area contains complete eye information such as the corners of the eyes. Using the expanded length as the benchmark, the parallel center line as the baseline, and the aspect ratio of 5:4, the images are cropped to ensure that all eye images meet the size requirements and maximize the effective area.

The human eye classifier is a deep learning model that has been pre-trained with a large number of human eye and non-human eye image samples. It is used to distinguish human eyes and non-human eyes in the input image, and perform eye region framing on images that are detected as human eyes in the input result. Select, and continue to perform subsequent image preprocessing on the frame selection results that meet the eye area range requirements, including scleral area detection, eye red area extraction, and eye red proportion calculation and analysis. For images whose eye area frame selection results do not meet the requirements, and for images whose input results are detected as non-human eyes, the algorithm terminates and gives corresponding prompt information. Among them, pre-training refers to using a large-scale data set to conduct initial training of the model before the target task, so that the model can learn some common features and representations. Eye area frame selection is based on the human eye determination result, using the label box to select the area determined to be the human eye, and returns the image coordinates of the frame selection area and the aspect ratio of the frame selection result. The frame selection result should contain complete eye information such as the inner and outer corners of the eyes, while minimizing other parts of the input image that are irrelevant to the human eye. The scope of the eye area requires that the aspect ratio of the eye area selected by the frame should not be less than 256×256, which facilitates subsequent steps such as scleral area detection and eye red area extraction. If the aspect ratio of the selected eye area is not less than 256×256, then the input eye image meets the requirements and you can continue with the subsequent steps. Otherwise, it will prompt that the human eye image is too small and require re-shooting.

Step two, sclera area detection.

The actual eye image obtained after cropping the input image is scaled to a size of 512×512, and after preprocessing, the improved U-Net++ model is used for scleral area detection. The model detection result is a binary image of the sclera area of the image to be inspected. The binary image and the zoom image are ANDed to obtain a color image of the sclera area.

Image scaling is to perform eye area cropping and preprocessing on the input image. After the image length is less than 512×512, super-resolution is performed. If the length is greater than 512×512, down-sampling is performed to reduce the loss of image details caused by scaling. Super-resolution is to increase the image resolution through interpolation method for target images whose actual resolution is lower than the required resolution, thereby improving the details and quality of the image. Downsampling is to average some pixels in the target image whose actual resolution is higher than the required resolution, thereby reducing the spatial resolution of the image.

Image preprocessing uses the color threshold method. If the overall color of the image is judged to be reddish, the R channel of the three RGB channels of the image is extracted and the scleral segmentation operation is continued. Otherwise, after using the histogram equalization method for image enhancement, the RGB three-channel color image is still used for the scleral segmentation operation.

The color threshold method calculates the average value of the red channel in the image to determine whether it is necessary to extract a single red channel from the image before performing scleral segmentation. When the average value of the color channels exceeds the set threshold, it is considered that the red single channel of the image needs to be extracted, otherwise it is not needed.

Histogram equalization obtains a histogram of pixel values by counting the frequency of occurrence of each pixel value in the image. After normalizing the histogram, calculates the cumulative distribution function (CDF) of the histogram, which is used to Represents the cumulative occurrence probability of each pixel value, calculates the mapping value of each pixel value in the new histogram according to the CDF, and maps the original pixel to the new pixel value. Use new pixel values to replace the original pixel values to obtain the equalized image. The cumulative distribution function CDF is expressed using the following formula:

Among them, x is the pixel value, and P(i) is the frequency of the pixel value i in the normalized image.

Histogram normalization is to normalize the frequency values in the histogram so that the frequency values are between 0 and 1. The normalization method used here is to divide each frequency value by the total number of pixels to ensure that the normalized frequency values sum to 1. Expressed using the following formula:

Among them, N(i) is the normalized frequency value, H(i) is the frequency value with pixel value i in the original histogram, and N _pixels is the total number of pixels. The normalized frequency value represents the relative probability of occurrence of pixel value i.

The three RGB channels refer to the three color channels of red (R), green (G), and blue (B) of the image. Each pixel of the image is composed of the values of the three color channels, which represent the intensity or intensity of red, green, and blue respectively. Brightness, the value of each color channel is usually an integer between 0 and 255. The higher the value, the greater the color component of the channel, and the brighter it is reflected on the single channel.

The U-Net++ model is a deep learning model for image segmentation tasks. It is an expansion and improvement of the classic U-Net model. U-Net++ enhances the expressive ability of the model by introducing a recursive network structure and improves image quality. Segmentation performance has been widely used in the field of medical image segmentation. The improved U-Net++ model in the present invention mainly adjusts the input and output parts of the original U-Net++ network, the convolution operation of each layer, and some structures between the upper and lower layers. In the input and output part, the input image is the image after the preprocessing step, the output image is a single-channel binary image, and the input and output image shapes are both 512×512. A normalization operation is added between each layer of convolution operation and ReLU activation function operation of U-Net++. Between the upper layer and the lower layer, a dropout operation is added, and an attention mechanism is added later, followed by a pooling operation. After improving according to the above operations, the accuracy can be improved by 2% to 3% on the test data.

A single-channel binary image refers to a binary image that contains only a single channel (grayscale channel). Each pixel has only two possible values: black and white. Typically a grayscale value of 0 is used to represent a black pixel, and a grayscale value of 255 (or 1) is used to represent a white pixel.

The full name of ReLU activation function is Rectified Linear Unit, and its Chinese name is linear rectification function, which is a commonly used activation function in neural networks. In a common sense, it refers to the slope function in mathematics.

Dropout is a strategy widely used in deep learning to solve the problem of model overfitting. Dropout solves the co-adaption problem and makes it possible to train wider networks.

The co-adaption problem refers to a co-adaptation problem. Some nodes in the network will have stronger representation capabilities than other nodes. As the network continues to train, nodes with stronger representation capabilities are continuously strengthened, while weaker ones The nodes continue to weaken until their contribution to the network is negligible. At this time, only some nodes in the network will be trained, which wastes the width and depth of the network, thus limiting the effectiveness of the model.

The attention mechanism is a special structure that people embed in machine learning models to automatically learn and calculate the contribution of input data to output data. By introducing the attention mechanism, the neural network can automatically learn and selectively focus on important information in the input, improving the performance and generalization ability of the model.

The pooling operation imitates the human visual system to reduce the dimensionality of the data, mainly to propose the key information of the area in a certain area.

Model detection is to use the trained and improved U-Net++ model to detect the input image. The input image needs to be detected by the human eye classifier in the first step. After cropping the human eye image, adjust the image size to 512× 512. As the input of the model, the model can accurately detect the sclera area in the input image and give a binary map of the detection image result.

Perform the AND operation on the binary image and the zoom image, set all pixels with a pixel value of 0 (black part) in the binary image, and set the corresponding pixel values in the zoom image to 0, and only retain the sclera area in the image to obtain the sclera area. picture.

Step 3: Extract the eye red area.

The color image of the sclera area obtained in the previous step is used as the input image. Image smoothing and adaptive histogram equalization are performed on the input image to enhance the image contrast. The B-COSFIRE filter is used on the processed image to extract blood streaks and enhance the image at the same time. The mask image is obtained from the processed color image, and the RGB color model of the color image is converted into a LAB color model. A suitable mask is regenerated according to the threshold value set for the L (brightness) component, and the low-brightness area (i.e., the black-gray background area) is regenerated. The mask is set to 0, and the remaining masks are set to 1. This can effectively prevent the intersection between black and color from being recognized as bloodshot by the algorithm. Use the B-COSFIRE filter to extract blood filaments from the processed image, and the extracted result image is a binary image of blood filaments.

Image smoothing uses the Gaussian filtering method to perform the convolution operation of the Gaussian kernel function on the image to achieve a smoothing effect, remove high-frequency noise in the image and smooth the image, and maintain the overall details and edges of the image.

Adaptive histogram equalization is to divide the original image into uniform small blocks or divide it according to a specific algorithm, perform histogram equalization on each small block, and perform brightness correction according to the histogram distribution of adjacent small blocks. , and finally recombine all the small patches into an enhanced image. Equalized images can improve local contrast by avoiding excessive amplification of noise in relatively uniform areas.

The RGB color model is a model used to describe colors, based on a combination of three basic color channels: red, green, and blue. The value of each color channel is usually between 0 and 255. 0 represents the minimum brightness and 255 represents the maximum brightness.

The LAB color model is a model used to describe color, consisting of two channels: brightness (L) and color opposition (A and B). The A channel represents the opposition from green to red, and the B channel represents the opposition from blue to The opposite nature of yellow.

B-COSFIRE filter is the abbreviation of Bar-selective shift filter response combination (Combination Of Shifted FIlterREsponses). B-COSFIRE filters achieve direction selectivity by computing a weighted geometric mean of the outputs of a set of Gaussian difference filters whose support regions are aligned in a linear fashion. Segmentation of ocular blood vessels is achieved by summing and thresholding the responses of two rotation-invariant B-COSFIRE filters.

Rotation invariance means that when processing images, the algorithm or method has the same recognition or analysis results for objects at different rotation angles. That is, no matter how the object is rotated, the algorithm can correctly identify or process it without being affected by the rotation.

The B-COSFIRE filter response function is:

In the above formula, σ is the standard deviation of the Gaussian function that determines the response range.

Step 4: Calculate the proportion of jealousy and judge the degree of jealousy.

Based on the above operation steps, the obtained bloodshot binary image and scleral area map are calculated respectively. The number of non-zero pixels in the bloodshot binary image and scleral area map is calculated, and then the ratio is calculated. The ratio between the two is the eye red ratio. Multiply the calculation result of the jealousy ratio by 100 and round down, so that the final result is within the range of [0,100]. Finally, the following score rating system is used to obtain the corresponding jealousy level, as shown in Table 1.

Among them, Degree represents the degree of jealousy, B _pixels represents the number of non-0 pixels in the bloodshot binary image, and S _pixels represents the number of non-0 pixels in the sclera area map.

Table 1 Degree of jealousy

Degree 81～100 61～80 41～60 21～40 0～20 grade A B C D E

The invention also discloses a non-contact automatic eye redness analysis device, which includes:

The eye image acquisition and processing module is used to use visible light imaging equipment to shoot human eyes from the front of the face, and perform human eye area detection and image preprocessing on the captured images to obtain preprocessed images;

The scleral area detection module is used to scale the preprocessed image to obtain a zoomed image. After preprocessing the zoomed image, use the U-Net++ model to detect the scleral area of the zoomed image to obtain a binary image of the scleral area of the image to be inspected. Perform operations on the binary image and the zoom image to obtain a color image of the sclera area;

The eye-red area extraction module is used to extract the color image of the sclera area, perform image smoothing and adaptive histogram equalization to enhance the image contrast, and obtain the image enhancement map, and then use the B-COSFIRE filter to filter the image enhancement map to obtain Filter the processing map, and at the same time convert the image enhancement map into a LAB color model to obtain the image mask. Finally, perform the AND operation on the filtering processing map and the image mask to obtain the bloodshot binary image;

The eye red proportion calculation module is used to calculate the number of non-zero pixels in the sclera area color image and the bloodshot binary image respectively, and then calculate the number of non-zero pixels in the bloodshot binary image and the sclera area color image. Perform a ratio calculation to get the jealousy ratio.

A computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the method for automatic analysis of the degree of contact envy of the present invention is implemented.

The above are only the preferred embodiments of the present invention. It should be pointed out that those of ordinary skill in the art can make several improvements and modifications without departing from the principles of the present invention. These improvements and modifications can also be made. should be regarded as the protection scope of the present invention.

Claims (10)

1. A non-contact automatic analysis method of eye redness degree, characterized by: including the following steps: Use visible light imaging equipment to shoot the human eyes from the front of the face, and perform human eye area detection and image preprocessing on the captured images to obtain preprocessed images; The preprocessed image is scaled to obtain a zoomed image. After preprocessing the zoomed image, the U-Net++ model is used to detect the scleral area, and a binary image of the scleral area of the image to be inspected is obtained. The binary image is combined with the zoomed image. Get a color map of the sclera area; Perform image smoothing and adaptive histogram equalization on the scleral area color map to enhance image contrast and obtain an image enhancement map. Then use the B-COSFIRE filter to filter the image enhancement map to obtain a filtered map. At the same time, the image is enhanced. The image is converted into a LAB color model, and the image mask is obtained. Finally, the filtered image and the image mask are ANDed to obtain the bloodshot binary image; Calculate the number of non-zero pixels in the color map of the sclera area and the binary blood map respectively, and then calculate the ratio of the number of non-zero pixels in the binary blood map and the color map of the sclera area to obtain the proportion of eye redness. . 2. A non-contact automatic eye redness analysis method according to claim 1, characterized in that: the visible light imaging device includes a smart phone and a camera, and the eye area imaging size needs to be no less than 256×256. 3. A kind of non-contact eye redness automatic analysis method according to claim 1, characterized in that: the human eye area detection includes the following steps: When detecting the captured image, use the trained classifier to search area by area starting from the upper left corner of the captured image, and use the similarity criterion to determine whether it is a human eye; If the judgment result is human eyes, select the range where the eye area is located and determine whether the eye area range is no less than 256×256. If so, proceed to the next step. If not, it will prompt that the human eye image is too small and require a new process. Photography; If the judgment result is not human eyes, it will prompt that the input image is wrong and require re-shooting; Among them, the training process of the human eye classifier is: using the deep learning method, first collect a large number of human eye and non-human eye samples from the Internet for pre-training, and obtain the deep learning model parameters to build the human eye classifier. 4. A kind of non-contact eye redness automatic analysis method according to claim 3, characterized in that: capturing images for preprocessing includes the following steps: According to the result of the eye area frame selection, the length and width of the human eye area are expanded by 1.2 times according to the original proportion, so that the final eye area selected by the frame contains complete eye information at the corner of the eye. The expanded length is used as the benchmark, and the parallel center line is the baseline. , cropped with an aspect ratio of 5:4 so that all eye images meet the size requirements and maximize the effective area. 5. A kind of non-contact eye redness automatic analysis method according to claim 1, characterized in that: the zoom image preprocessing includes the following steps: using the color threshold method, if the overall color of the image is judged to be reddish, then extract the three RGB channels of the image. In the R channel, continue the scleral segmentation operation. Otherwise, after using the histogram equalization method for image enhancement, the RGB three-channel color image is still used for the scleral segmentation operation. 6. A kind of non-contact automatic redness analysis method according to claim 1, characterized in that: the U-Net++ model is an improved U-Net++ model, the input image is a zoom image, and the output image is a single-channel binary image. , the input and output image shapes are both 512×512; a normalization operation is added between each layer of convolution operation and ReLU activation function operation of U-Net++; a dropout operation is added between the upper layer and the lower layer, and The attention mechanism is then added, followed by pooling operations. 7. A kind of non-contact automatic analysis method of eye redness degree according to claim 1, characterized in that: the operation of the binary image and the zoom image is: all the pixel points with a pixel value of 0 in the binary image are corresponding to The pixel value in the zoomed image is set to 0, and only the scleral area in the image is retained to obtain the scleral area map. 8. A kind of non-contact automatic analysis method of eye redness degree according to claim 1, characterized in that: the specific calculation method of eye redness proportion is as follows: Among them, Degree represents the proportion of jealousy, B _pixels represents the number of non-0 pixels in the bloodshot binary image, and S _pixels represents the number of non-0 pixels in the sclera area map. 9. A non-contact automatic analysis method of eye redness degree according to claim 8, characterized in that: multiply the calculation result of eye redness proportion by 100 and then round down, so that the final result is within the range of [0,100], and finally use Points grading system to obtain the corresponding jealousy level. 10. A non-contact automatic eye redness analysis device, characterized by: including: The eye image acquisition and processing module is used to use visible light imaging equipment to shoot human eyes from the front of the face, and perform human eye area detection and image preprocessing on the captured images to obtain preprocessed images; The scleral area detection module is used to scale the preprocessed image to obtain a zoomed image. After preprocessing the zoomed image, use the U-Net++ model to detect the scleral area of the zoomed image to obtain a binary image of the scleral area of the image to be inspected. Perform operations on the binary image and the zoom image to obtain a color image of the sclera area; The eye-red area extraction module is used to extract the color image of the sclera area, perform image smoothing and adaptive histogram equalization to enhance the image contrast, and obtain the image enhancement map, and then use the B-COSFIRE filter to filter the image enhancement map to obtain Filter the processing map, and at the same time convert the image enhancement map into a LAB color model to obtain the image mask. Finally, perform the AND operation on the filtering processing map and the image mask to obtain the bloodshot binary image; The eye red proportion calculation module is used to calculate the number of non-zero pixels in the sclera area color image and the bloodshot binary image respectively, and then calculate the number of non-zero pixels in the bloodshot binary image and the sclera area color image. Perform a ratio calculation to get the jealousy ratio.