WO2021189770A1 - 基于人工智能的图像增强处理方法、装置、设备及介质 - Google Patents

基于人工智能的图像增强处理方法、装置、设备及介质 Download PDF

Info

Publication number
WO2021189770A1
WO2021189770A1 PCT/CN2020/112331 CN2020112331W WO2021189770A1 WO 2021189770 A1 WO2021189770 A1 WO 2021189770A1 CN 2020112331 W CN2020112331 W CN 2020112331W WO 2021189770 A1 WO2021189770 A1 WO 2021189770A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
processed
standard
pixel
statistical
Prior art date
Application number
PCT/CN2020/112331
Other languages
English (en)
French (fr)
Inventor
范栋轶
王瑞
王立龙
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Priority to US17/613,482 priority Critical patent/US20220245803A1/en
Publication of WO2021189770A1 publication Critical patent/WO2021189770A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/40—Image enhancement or restoration using histogram techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/60—Image enhancement or restoration using machine learning, e.g. neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/13—Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/143—Segmentation; Edge detection involving probabilistic approaches, e.g. Markov random field [MRF] modelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/174—Segmentation; Edge detection involving the use of two or more images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/60—Analysis of geometric attributes
    • G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20112—Image segmentation details
    • G06T2207/20132—Image cropping

Definitions

  • This application relates to the field of image processing, and in particular to an artificial intelligence-based image enhancement processing method, device, equipment and medium.
  • image recognition technology With the continuous development of image recognition technology, more and more fields apply image recognition technology to improve work efficiency. Especially in the medical field, image recognition technology is used to automatically recognize images, so that the recognition results can quickly identify the lesion.
  • the inventor realizes that due to the shooting quality of different equipment, the collection environment and the operating proficiency of the shooting physician, the captured images have non-target features, and there are large differences in the distribution of image pixel values, resulting in subsequent automatic recognition of images. There is more interference, and there are errors in the recognition results.
  • An image enhancement processing method based on artificial intelligence including:
  • the edge detection algorithm is used to perform edge detection on the original feature image, obtain the original gradient map, obtain a statistical loop based on the original feature map, and perform iterative processing on the statistical loop.
  • the statistical loop intersects the original gradient map, it is based on the statistical loop
  • An image enhancement processing device based on artificial intelligence including:
  • the original feature image obtaining module is used to obtain an initial image, preprocess the initial image, and obtain an original feature image containing the target feature.
  • the image acquisition module to be processed is used to perform edge detection on the original feature image by using an edge detection algorithm, obtain an original gradient map, obtain a statistical ring based on the original feature map, and perform iterative processing on the statistical ring.
  • the original gradient map intersects, the original feature image is captured based on the inner diameter of the statistical ring to obtain the image to be processed, and the image parameters of the image to be processed are determined:
  • the standard image parameter acquisition module is used to acquire a standard image corresponding to the target feature, determine a standard area corresponding to the standard image, and obtain a standard image parameter corresponding to the standard area.
  • the migration image acquisition module is configured to perform migration processing on the to-be-processed image according to the to-be-processed parameter and the standard image parameter to acquire the migration image.
  • the target enhanced image acquisition module is configured to perform a restricted contrast adaptive histogram equalization process on the migration image to acquire a target enhanced image.
  • a computer device includes a memory, a processor, and a computer program that is stored in the memory and can run on the processor, and the processor implements the following steps when the processor executes the computer program:
  • the edge detection algorithm is used to perform edge detection on the original feature image, obtain the original gradient map, obtain a statistical loop based on the original feature map, and perform iterative processing on the statistical loop.
  • the statistical loop intersects the original gradient map, it is based on the statistical loop
  • a computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:
  • the edge detection algorithm is used to perform edge detection on the original feature image, obtain the original gradient map, obtain a statistical loop based on the original feature map, and perform iterative processing on the statistical loop.
  • the statistical loop intersects the original gradient map, it is based on the statistical loop
  • FIG. 1 is a schematic diagram of an application environment of an image enhancement processing method based on artificial intelligence in an embodiment of the present application
  • FIG. 2 is a flowchart of an image enhancement processing method based on artificial intelligence in an embodiment of the present application
  • FIG. 3 is another flowchart of an image enhancement processing method based on artificial intelligence in an embodiment of the present application
  • FIG. 4 is another flowchart of an image enhancement processing method based on artificial intelligence in an embodiment of the present application.
  • FIG. 5 is another flowchart of an image enhancement processing method based on artificial intelligence in an embodiment of the present application
  • FIG. 6 is another flowchart of an image enhancement processing method based on artificial intelligence in an embodiment of the present application.
  • FIG. 7 is another flowchart of an image enhancement processing method based on artificial intelligence in an embodiment of the present application.
  • FIG. 8 is a functional block diagram of an image enhancement processing device based on artificial intelligence in an embodiment of the present application.
  • Fig. 9 is a schematic diagram of a computer device in an embodiment of the present application.
  • the artificial intelligence-based image enhancement processing method can be applied to the application environment as shown in FIG. 1.
  • the artificial intelligence-based image enhancement processing method is applied in an artificial intelligence-based image enhancement processing system.
  • the artificial intelligence-based image enhancement processing system includes a client and a server as shown in FIG.
  • the network communicates to achieve accurate segmentation of target features and high-quality images.
  • the client is also called the client, which refers to the program that corresponds to the server and provides local services to the client.
  • the client can be installed on, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices.
  • the server can be implemented as an independent server or a server cluster composed of multiple servers.
  • an image enhancement processing method based on artificial intelligence is provided. Taking the method applied to the server in FIG. 1 as an example for description, the method includes the following steps:
  • S201 Obtain an initial image, perform preprocessing on the initial image, and acquire an original feature image containing the target feature.
  • the target feature refers to human tissue that needs to be identified and analyzed by computer vision technology in order to subsequently determine the user's symptoms
  • the target feature can be eyes, mouth, etc.
  • the initial image is an image of the facial features of the human face captured by the image capturing device.
  • the image capturing device may be a mobile phone, a medical image capturing device, and the like.
  • the original feature image refers to cutting the initial image with the length and width of the target feature as the crop size to obtain a rectangular image centered on the target feature. Understandably, the original feature image contains non-target features, that is, the original feature image Contains noise.
  • the target feature of the initial image is intercepted, so that the target feature can be processed subsequently, the processing area of the subsequent image processing is reduced, and the noise interference is reduced.
  • S202 Use an edge detection algorithm to perform edge detection on the original feature image, obtain the original gradient map, obtain the statistical ring based on the original feature map, and perform iterative processing on the statistical ring.
  • the original The feature image is screenshot processed to obtain the image to be processed, and the image parameters of the image to be processed are determined.
  • the edge detection algorithm is an algorithm used to detect the region with the most obvious gray-scale change on the image.
  • the edge detection algorithm includes but is not limited to the Sobel algorithm, the Laplace edge detection algorithm, and the Canny edge detection algorithm.
  • the original gradient map is an image that represents the target feature edge of the original feature image.
  • the statistical ring is a ring with an inner diameter larger than the original gradient map, that is, the original gradient map is located in the statistical ring.
  • the iterative gradient value refers to the gradient value of the area within the statistical ring (that is, the area corresponding to the difference between the outer diameter of the statistical ring and the inner diameter of the statistical ring). Specifically, the area inside the statistical ring is masked with a mask of 1, and the area outside the statistical ring is masked with a mask of 0, the original gradient map and the statistical ring are multiplied, and the mask sum of the area in the statistical ring is calculated. Get the iterative gradient value. It should be noted that the center of the statistical ring coincides with the center of the original gradient map.
  • the image to be processed is an image that only contains target features.
  • the edge detection algorithm is used to process the original feature map to obtain the original gradient map, and then the precise location of the target feature is determined according to the statistical loop and the original gradient map to obtain the to-be-processed image containing only the target feature, so as to Computer vision technology can recognize and analyze accurately to-be-processed images to ensure that the recognition and analysis results are more accurate.
  • the image parameter to be processed is the pixel value corresponding to the image to be processed, and the image parameter to be processed may be the average value of the pixels to be processed and the variance of the pixels to be processed.
  • the edges may be continuous or discontinuous or inconsistent with the center distance of the original gradient map. If the original gradient map is directly used to compare the original features
  • the target feature is obtained by segmenting the graph, and the target feature is often inaccurate, and there are problems of incomplete or too many non-target features.
  • the obtained recognition result has errors.
  • the edge detection algorithm is used to detect the original feature image to obtain the edge of the target feature, that is, the original gradient map is obtained, the product operation is performed on the statistical loop and the original gradient map, and the inner diameter of the statistical loop is reduced iteratively Calculate the mask sum of the area within the statistical ring in the iterative process to form the iterative gradient value. Since the original gradient map has edge gradient values, if the inner diameter of the statistical ring does not intersect the original gradient map, iterate The gradient value is 0. If the inner diameter of the statistical ring intersects the original gradient graph, the iterative gradient value increases suddenly. When the iterative gradient value suddenly increases, it means that the statistical ring intersects the original gradient map.
  • S203 Obtain a standard image corresponding to the target feature, determine a standard area corresponding to the standard image, and obtain standard image parameters corresponding to the standard area.
  • the standard image refers to an image with high brightness, contrast, and color space quality. Using computer vision technology to recognize the standard image can obtain a better recognition effect.
  • the standard image parameter is the pixel parameter corresponding to the target feature in the standard image.
  • the standard image parameters include standard pixel average and standard pixel variance.
  • the standard image parameters corresponding to the standard area are acquired, so that the image to be processed can be transferred according to the standard image parameters to a transferred image that is easy to recognize.
  • S204 Perform migration processing on the to-be-processed image according to the to-be-processed parameter and the standard image parameter to obtain the migrated image.
  • the migration image refers to an image obtained by performing migration processing on the image to be processed, and the migration image has better brightness and contrast.
  • the pixel to be processed at any point on the image to be processed is acquired, so as to calculate the pixel to be processed according to the pixel to be processed, the parameter to be processed, and the standard image parameter to obtain the migration pixel corresponding to the pixel to be processed, and the migration image is formed according to the migration pixel.
  • the image to be processed is subjected to migration processing, so that the brightness and contrast of the image to be processed are consistent with the brightness and contrast of the standard image, so that the image to be processed is standardized.
  • S205 Perform a restricted contrast adaptive histogram equalization process on the migration image to obtain a target enhanced image.
  • the limited contrast adaptive histogram equalization is a computer image processing technology to improve the contrast of the image, which can improve the contrast of the local details of the migrated image, making the image easier to be recognized, and improving the effect of image enhancement processing based on artificial intelligence.
  • block processing is performed on the calculated migration image to obtain multiple block images to be processed, the block histogram of each block image to be processed is calculated, and the amplitude of each block histogram is higher than the preset value.
  • the thresholded part is limited to obtain a cropped histogram, and the cropped histogram is histogram equalized to obtain the target enhanced image to ensure a better recognition effect.
  • the artificial intelligence-based image enhancement processing method acquires an initial image, preprocesses the initial image, and acquires the original feature image containing the target feature, so that the target feature can be subsequently processed and the subsequent image processing is reduced. Area to reduce noise interference.
  • the edge detection algorithm is used to perform edge detection on the original feature image, obtain the original gradient map, obtain the statistical loop based on the original feature map, and perform iterative processing on the statistical loop. When the statistical loop intersects the original gradient map, the original feature image is based on the inner diameter of the statistical loop.
  • the image parameters are used for subsequent migration processing of the image to be processed to obtain an image with better image quality, so as to improve the effect of image processing.
  • Obtain the standard image corresponding to the target feature determine the standard area corresponding to the standard image, and obtain the standard image parameters corresponding to the standard area, so that the image to be processed can be transferred according to the standard image parameters to a transferred image that is easy to recognize.
  • the to-be-processed image is transferred and the transferred image is obtained, so that the brightness and contrast of the to-be-processed image are consistent with the brightness and contrast of the standard image, so that the to-be-processed image is standardized.
  • the edge detection algorithm is used to perform edge detection on the original feature image, obtain the original gradient map, obtain the statistical loop based on the original feature map, and perform iterative processing on the statistical loop.
  • the original feature image is screenshot processed to obtain the image to be processed, including:
  • S301 Calculate the horizontal gradient in the horizontal direction and the vertical gradient in the vertical direction in the original feature image by using the Sobel algorithm.
  • the Sobel algorithm is a discrete difference algorithm, which combines Gaussian smoothing and differential derivation to calculate the gray scale approximation of the image brightness function. Using Sobel operator at any point of the image will generate the corresponding gray vector or its normal vector.
  • the Sobel algorithm is used to obtain the horizontal gradient and the vertical gradient of the original feature image, so as to obtain the approximate position of the target feature edge in the original feature image according to the horizontal gradient and the vertical gradient.
  • S302 Perform weighting processing on the horizontal gradient and the vertical gradient to obtain an original gradient map.
  • the horizontal gradient and the vertical gradient are weighted according to the horizontal gradient and the vertical gradient to obtain the original gradient map, thereby obtaining the approximate position of the target feature edge, so as to subsequently obtain the precise position of the target feature based on the original gradient map.
  • the edges of the original gradient map obtained by the Sobel algorithm are relatively blurry, and may be continuous or discontinuous or inconsistent with the center distance of the original gradient map.
  • S303 Obtain a statistical ring according to the length of the long side of the original feature image and the preset spacing.
  • the preset distance is a preset distance, which is used to determine the statistical ring according to the length of the long side of the original feature image (the long side refers to the longer side in the original feature image) and the preset distance. Specifically, the center of the original feature map is taken as the center of the statistical ring, the length of the long side of the original feature map is taken as the outer diameter of the statistical ring, and the difference between the outer diameter and the preset spacing is taken as the inner diameter of the statistical ring to obtain the statistical ring.
  • the statistical loop can be used to obtain the precise position of the target feature in the original feature image, so as to accurately segment the image to be processed.
  • S304 Perform an iterative process of reducing the inner diameter of the statistical loop to obtain an iterative gradient value corresponding to the statistical loop.
  • the inner diameter of the statistical loop is reduced iteratively.
  • the iterative gradient value of the statistical loop is required to determine whether the statistical loop intersects the original gradient map to accurately segment from the original feature map. The precise location of the target feature.
  • the inner diameter of the statistical ring intersects the original gradient graph. Therefore, the center of the original gradient is taken as a circle, the inner diameter of the statistical ring is taken as the diameter, and the screenshot is taken as Obtain the image to be processed, so as to accurately determine the location of the target feature, ensure that the subsequent image processing is targeted, and reduce interference.
  • the artificial intelligence-based image enhancement processing method provided in this embodiment adopts the Sobel algorithm to calculate the horizontal gradient in the horizontal direction and the vertical gradient in the vertical direction in the original feature image, weights the horizontal gradient and the vertical gradient, and obtains the original gradient map, Get the approximate position of the target feature edge in the original feature image.
  • the statistical ring is obtained, the inner diameter of the statistical ring is reduced iteratively, and the iterative gradient value corresponding to the statistical ring is obtained, so as to determine whether the statistical ring intersects with the original gradient graph to accurately The precise position of the target feature is segmented from the original feature map.
  • the inner diameter of the statistical loop intersects the original gradient map, and the inner circle of the statistical loop is determined as the image to be processed, so as to accurately determine the position of the target feature and ensure that the subsequent image processing is targeted Sex, reduce interference.
  • step S204 that is, performing migration processing on the image to be processed according to the parameters to be processed and the standard image parameters, and obtaining the migrated image includes:
  • S401 Acquire each pixel to be processed of the image to be processed.
  • the pixels to be processed refer to pixels in the image to be processed.
  • S402 Use the migration formula to process the pixel to be processed, the parameter to be processed, and the standard image parameter, and obtain the migrated pixel corresponding to the pixel to be processed.
  • the shifted pixel is a processed pixel of the pixel to be processed, and one shifted pixel corresponds to one to-be-processed pixel to form a shifted image corresponding to the to-be-processed image.
  • each pixel to be processed in the image to be processed is acquired, and the pixel to be processed, the parameters to be processed, and the standard image parameters are substituted into the migration formula to obtain the migration pixel corresponding to the pixel to be processed, which provides technical support for the subsequent formation of the migration image .
  • the migration image is formed according to the migration pixels, which improves the brightness of the migration image and ensures a better color space, reduces interference, and ensures that the recognition effect of subsequent image recognition using computer vision technology is better.
  • the artificial intelligence-based image enhancement processing method acquires each pixel to be processed of the image to be processed.
  • the migration formula is adopted to process the pixels to be processed, the parameters to be processed and the standard image parameters, to obtain the migration pixels corresponding to the pixels to be processed, and to provide technical support for the subsequent formation of the migration image.
  • the shifted image is formed, which improves the brightness of the shifted image and ensures a better color space, reduces interference, and ensures that the subsequent use of computer vision technology for image recognition has a better recognition effect.
  • the image parameters to be processed include the average value of the pixels to be processed and the variance of the pixels to be processed.
  • Standard image parameters include standard pixel average and standard pixel variance.
  • Step S402 that is, the pixel to be processed, the parameter to be processed and the standard image parameter are processed by the migration formula, and the migration pixel corresponding to the pixel to be processed is obtained, which includes:
  • the average value of pixels to be processed refers to the average value of all pixels in the image to be processed.
  • the variance of the pixels to be processed refers to the variance of all pixels in the image to be processed.
  • the standard pixel average value refers to the average value of all pixels in the standard area.
  • the standard pixel variance refers to the variance of all pixels in the standard area.
  • the migration formula is:
  • img output (img input -mean input )/(std input *std template )+mean template
  • img output is the shifted pixel
  • img input is the pixel to be processed
  • mean input is the average value of the pixels to be processed
  • std input is the variance of the pixels to be processed
  • mean template is the average value of standard pixels
  • std template is the standard pixel variance.
  • the migration formula is used to realize the migration processing of the image to be processed to form the migration image, so as to ensure that the brightness and color space of the image to be migrated are better, and the recognition effect of the computer vision technology is better.
  • step S205 that is, performing a restricted contrast adaptive histogram equalization process on the migration image to obtain a target enhanced image includes:
  • the channel image refers to the image corresponding to the L channel.
  • the channel image contains brightness information for processing according to the brightness information to improve the local detail contrast information of the migrated image and ensure that the target enhanced image has higher definition, so that the image
  • the enhancement processing effect is better, therefore, the subsequent computer vision technology recognition effect is better.
  • the migration image is converted into the Lab channel, and only the L channel is processed, so as to process according to the brightness information, improve the local detail contrast information of the migration image, and avoid the use of RGB three-channel separate processing to cause serious color cast.
  • S502 Perform block processing on the channel image to obtain a block image to be processed.
  • the channel image is uniformly divided into 8*8 blocks, and 64 to-be-processed block images are obtained, so that each block-to-be-processed block image is processed separately to improve the local detail contrast information of the block-to-be-processed image and avoid processing.
  • the processing of the block image leads to the problem that the entire gray dynamic range is not fully utilized, and the local contrast cannot be effectively improved.
  • S503 Use a histogram function to process the block graph to be processed, and obtain a block histogram corresponding to the block graph to be processed.
  • the histogram function is a function that converts the block graph to be processed into a block histogram.
  • the block histogram can be quickly obtained by the histogram function, so as to perform gray scale transformation according to the pixel distribution of the block histogram, so as to achieve the purpose of improving the contrast of the image.
  • S504 Crop the block histogram based on the preset threshold, obtain the cropped histogram, equalize the cropped histogram, and obtain the target enhanced image.
  • the preset threshold is a value used for segmenting the histogram for cropping.
  • the mapping function is obtained by cropping the block histogram, and the value of each pixel in the block image to be processed is obtained by bilinear interpolation from the mapping function values of the four surrounding block histograms. According to the mapping function value, the pixels of the block image to be processed are transformed to improve the local detail contrast information, thereby obtaining the target enhanced image.
  • step S504 that is, cropping the block histogram based on a preset threshold to obtain the cropped histogram, includes:
  • S602 Evenly distribute the gray values higher than the preset threshold to all the block histograms, and obtain the cropped histograms.
  • the artificial intelligence-based image enhancement processing method before step S201, that is, before acquiring the original feature image containing the target feature, the artificial intelligence-based image enhancement processing method further includes:
  • S701 Acquire a verification image and at least two candidate images corresponding to the verification image.
  • the image to be selected refers to a high-quality template image collected in advance, and the image to be selected is screened to determine the image with the best quality as the standard image.
  • the verification image is an image used to verify the quality of the image to be selected for verification.
  • S702 Perform steps S201-S205 to process the verification image, and obtain an image to be tested corresponding to the verification image.
  • the image to be tested is an image obtained after the verification image is enhanced according to the image to be selected, and is used to verify the quality of the image to be selected.
  • the verification images are a1, a2, and a3; the candidate images are b1, b2, b3, and b4; the verification images a1, a2, and a3 are migrated according to the candidate image b1; the verification images a1, Perform migration processing on a2 and a3; perform migration processing on the verification images a1, a2, and a3 based on the candidate image b3; perform migration processing on the verification images a1, a2, and a3 based on the candidate image b4 to obtain the migration image, and perform migration processing on the migration image Limit the contrast adaptive histogram equalization process, and obtain the test images c11, c21, and c31 corresponding to the candidate image b1; the test images c12, c22, and c32 corresponding to the candidate image b2; and the candidate image corresponding to the candidate image b3 Test images c13, c23, and c33; test images c14, c24, and c34 corresponding to the candidate image
  • S703 Input the image to be tested into the MaskRCNN model generated based on the selected image corresponding to the image to be tested for detection, and obtain the detection result.
  • the detection result is the result of using the MaskRCNN model to scan the image to be selected, and generating the bounding box and mask of the target feature in the image to be selected, so as to filter the image to be selected.
  • the detection result indicates the proportion of the target feature that is selected by the MaskRCNN model in the image to be selected.
  • the detection result can be 100%, that is, the MaskRCNN model can accurately detect the target feature in the candidate image, that is, the detection result is accurate; or the detection result can be 50%, that is, the MaskRCNN model can detect the candidate image 50% of the target feature in, that is, the detection result is wrong.
  • S704 Determine the detection accuracy of each candidate image according to the detection result, and determine the candidate image with the highest detection accuracy as the standard image.
  • the detection accuracy is the accuracy used to indicate whether the detection result is accurate, that is, the detection accuracy is the number of accurate detection results divided by the total number of detection results.
  • the statistics are based on the detection accuracy corresponding to the image to be tested processed for each image to be selected, and the candidate image with the highest detection accuracy is selected and determined as the standard image, so that the image to be processed provides the highest quality reference image.
  • a verification image and at least two candidate images corresponding to the verification image are acquired; steps S201-S205 are executed to process the verification image, and the test image corresponding to the verification image is acquired to verify the quality of the candidate image.
  • an artificial intelligence-based image enhancement processing device corresponds to the artificial intelligence-based image enhancement processing method in the foregoing embodiment in a one-to-one correspondence.
  • the artificial intelligence-based image enhancement processing device includes an original feature image acquisition module 801, a to-be-processed image acquisition module 802, a standard image parameter acquisition module 803, a migration image acquisition module 804, and a target enhanced image acquisition module 805.
  • the detailed description of each functional module is as follows:
  • the original feature image obtaining module 801 is used to obtain an initial image, preprocess the initial image, and obtain an original feature image containing the target feature.
  • the to-be-processed image acquisition module 802 is used to use edge detection algorithms to perform edge detection on the original feature image, obtain the original gradient map, obtain the statistical loop based on the original feature map, and perform iterative processing on the statistical loop.
  • edge detection algorithms to perform edge detection on the original feature image
  • obtain the original gradient map obtain the statistical loop based on the original feature map
  • the statistical loop intersects the original gradient map
  • the standard image parameter acquisition module 803 is configured to acquire a standard image corresponding to the target feature, determine a standard area corresponding to the standard image, and obtain a standard image parameter corresponding to the standard area.
  • the migration image acquisition module 804 is configured to perform migration processing on the to-be-processed image according to the to-be-processed parameters and the standard image parameters to acquire the migrated image.
  • the target enhanced image acquisition module 805 is configured to perform limited contrast adaptive histogram equalization processing on the migration image to acquire the target enhanced image.
  • the to-be-processed image acquisition module 802 includes: a gradient calculation unit, an original gradient map acquisition unit, a statistical loop acquisition unit, an iterative gradient value acquisition unit, and a to-be-processed image acquisition unit.
  • the gradient calculation unit is used to calculate the horizontal gradient in the horizontal direction and the vertical gradient in the vertical direction in the original feature image using the Sobel algorithm.
  • the original gradient map acquisition unit is used to perform weighting processing on the horizontal gradient and the vertical gradient to acquire the original gradient map.
  • the statistical ring obtaining unit is used to obtain the statistical ring according to the length of the long side of the original feature image and the preset spacing.
  • the iterative gradient value acquisition unit is used to reduce the inner diameter of the statistical loop iteratively, and obtain the iterative gradient value corresponding to the statistical loop.
  • the to-be-processed image acquisition unit is configured to, if the iterative gradient value is greater than the preset gradient value, the inner diameter of the statistical ring intersects the original gradient map, and the inner circle of the statistical ring is determined as the image to be processed.
  • the migration image acquisition module 804 includes: a pixel acquisition unit to be processed, a migration pixel acquisition unit, and a migration image acquisition unit.
  • the to-be-processed pixel acquiring unit is used to acquire each to-be-processed pixel of the to-be-processed image.
  • the migration pixel acquiring unit is used to process the pixel to be processed, the parameter to be processed, and the standard image parameter using a migration formula, and to acquire the migration pixel corresponding to the pixel to be processed.
  • the migration image acquisition unit is used to form a migration image based on the migration pixels.
  • the image parameters to be processed include the average value of the pixels to be processed and the variance of the pixels to be processed.
  • Standard image parameters include standard pixel average and standard pixel variance.
  • the migration pixel acquisition unit includes: a migration pixel calculation unit.
  • the target enhanced image acquisition module 805 includes: a channel image acquisition unit, a block image acquisition unit to be processed, a block histogram acquisition unit, and an equalization processing unit.
  • the channel image acquisition unit is configured to acquire the corresponding channel image based on the migration image.
  • the to-be-processed block image acquisition unit is used to perform block processing on the channel image to obtain the to-be-processed block image.
  • the block histogram obtaining unit is used to process the block graph to be processed by using the histogram function, and obtain the block histogram corresponding to the block graph to be processed.
  • the equalization processing unit is used to crop the block histogram based on the preset threshold, obtain the cropped histogram, equalize the cropped histogram, and obtain the target enhanced image.
  • the equalization processing unit includes:
  • the gray value obtaining unit is used to obtain the gray value corresponding to the block image to be processed.
  • the crop histogram obtaining unit is used to evenly distribute the gray value higher than the preset threshold to all the block histograms to obtain the crop histogram.
  • the artificial intelligence-based image enhancement processing method device further includes: a candidate image acquisition module, a test image acquisition module, a detection result acquisition module, and a standard image acquisition module.
  • the candidate image acquisition module is used to acquire a verification image and at least two candidate images corresponding to the verification image.
  • the test image acquisition module is configured to perform the steps of claim 1 to process the verification image, and obtain the test image corresponding to the verification image.
  • the detection result acquisition module is used to input the image to be tested into the MaskRCNN model generated based on the selected image corresponding to the image to be tested for detection, and obtain the detection result.
  • the standard image acquisition module is used to determine the detection accuracy of each candidate image according to the detection result, and determine the candidate image with the highest detection accuracy as the standard image.
  • the various modules in the above artificial intelligence-based image enhancement processing device can be implemented in whole or in part by software, hardware, and a combination thereof.
  • the above-mentioned modules may be embedded in the form of hardware or independent of the processor in the computer equipment, or may be stored in the memory of the computer equipment in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.
  • a computer device is provided.
  • the computer device may be a server, and its internal structure diagram may be as shown in FIG. 9.
  • the computer equipment includes a processor, a memory, a network interface, and a database connected through a system bus.
  • the processor of the computer device is used to provide calculation and control capabilities.
  • the memory of the computer device includes a non-volatile or volatile storage medium and internal memory.
  • the non-volatile or volatile storage medium stores an operating system, a computer program, and a database.
  • the internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile or volatile storage medium.
  • the database of the computer equipment is used to store the target enhanced image.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • the computer program is executed by the processor to realize an artificial intelligence-based image enhancement processing method.
  • a computer device including a memory, a processor, and a computer program stored in the memory and capable of running on the processor.
  • the processor executes the computer program to implement the artificial intelligence-based image in the above embodiment.
  • the steps of the enhancement processing method such as steps S201-S205 shown in FIG. 2, or the steps shown in FIG. 3 to FIG. 7, are not repeated here in order to avoid repetition.
  • the function of each module/unit in the embodiment of the artificial intelligence-based image enhancement processing device is realized, such as the original feature image acquisition module 801, the to-be-processed image acquisition module 802, and the image enhancement processing device shown in FIG.
  • the functions of the standard image parameter acquisition module 803, the migration image acquisition module 804, and the target enhanced image acquisition module 80 are not repeated here in order to avoid repetition.
  • a computer-readable storage medium is provided.
  • the computer-readable storage medium may be volatile or non-volatile.
  • the computer-readable storage medium stores a computer program, and the computer program is
  • the processor implements the steps of the artificial intelligence-based image enhancement processing method in the foregoing embodiment when executed, such as steps S201-S205 shown in FIG. 2, or the steps shown in FIG. 3 to FIG. Go into details.
  • the processor executes the computer program, the function of each module/unit in the embodiment of the image enhancement processing device based on artificial intelligence is realized, such as the original feature image acquisition module 801, the to-be-processed image acquisition module 802, and the image enhancement processing device shown in FIG.
  • the functions of the standard image parameter acquisition module 803, the migration image acquisition module 804, and the target enhanced image acquisition module 80 are not repeated here in order to avoid repetition.
  • Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Software Systems (AREA)
  • Probability & Statistics with Applications (AREA)
  • Geometry (AREA)
  • Image Processing (AREA)
  • Image Analysis (AREA)

Abstract

一种基于人工智能的图像增强处理方法、装置、设备及介质，方法通过获取初始图像，对初始图像进行预处理，获取包含目标特征的原始特征图像（S201）；采用边缘检测算法对原始特征图像进行边缘检测，获取原始梯度图，基于原始特征图像获取统计环，对统计环进行迭代处理，当统计环与原始梯度图相交则基于统计环的内径对原始特征图像进行截图处理，得到待处理图像，并确定待处理图像的待处理图像参数（S202）；获取与目标特征相对应的标准图像，确定标准图像对应的标准区域，并获取标准区域对应的标准图像参数（S203）；根据待处理图像参数和标准图像参数，对待处理图像进行迁移处理，获取迁移图像（S204）；对迁移图像进行限制对比度自适应直方图均衡处理，获取目标增强图像（S205）。

Description

基于人工智能的图像增强处理方法、装置、设备及介质
本申请要求于2020年7月31日提交中国专利局、申请号为CN202010763141.4、名称为“基于人工智能的图像增强处理方法、装置、设备及介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及图像处理领域，尤其涉及一种基于人工智能的图像增强处理方法、装置、设备及介质。
背景技术
随着图像识别技术的不断发展，越来越多的领域应用图像识别技术以提高工作效率。特别是应用在医学领域中，采用图像识别技术对图像进行自动识别，以便识别结果快速地确定病灶。
发明人意识到受制于不同的设备拍摄质量以及采集环境和拍摄医师操作熟练度的不同，拍摄得到的图像存在非目标特征，图像像素值分布存在较大差异，导致后续对图像进行自动识别时存在较多干扰，识别结果存在误差。
发明内容
一种基于人工智能的图像增强处理方法，包括：
获取初始图像，对所述初始图像进行预处理，获取包含目标特征的原始特征图像；
采用边缘检测算法对所述原始特征图像进行边缘检测，获取原始梯度图，基于所述原始特征图获取统计环，对所述统计环进行迭代处理，当统计环与原始梯度图相交则基于统计环的内径对所述原始特征图像进行截图处理，得到待处理图像，并确定所述待处理图像的待处理图像参数：
获取与所述目标特征相对应的标准图像，确定所述标准图像对应的标准区域，并获取所述标准区域对应的标准图像参数；
根据待处理参数和所述标准图像参数，对所述待处理图像进行迁移处理，获取迁移图像；
对所述迁移图像进行限制对比度自适应直方图均衡处理，获取目标增强图像。
一种基于人工智能的图像增强处理装置，包括：
原始特征图像获取模块，用于获取初始图像，对所述初始图像进行预处理，获取包含目标特征的原始特征图像。
待处理图像获取模块，用于采用边缘检测算法对所述原始特征图像进行边缘检测，获取原始梯度图，基于所述原始特征图获取统计环，对所述统计环进行迭代处理，当统计环与原始梯度图相交则基于统计环的内径对所述原始特征图像进行截图处理，得到待处理图像，并确定所述待处理图像的待处理图像参数：
标准图像参数获取模块，用于获取与所述目标特征相对应的标准图像，确定所述标准图像对应的标准区域，并获取所述标准区域对应的标准图像参数。
迁移图像获取模块，用于根据待处理参数和所述标准图像参数，对所述待处理图像进行迁移处理，获取迁移图像。
目标增强图像获取模块，用于对所述迁移图像进行限制对比度自适应直方图均衡处理，获取目标增强图像。
一种计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现如下步骤：
获取初始图像，对所述初始图像进行预处理，获取包含目标特征的原始特征图像；
采用边缘检测算法对所述原始特征图像进行边缘检测，获取原始梯度图，基于所述原始特征图获取统计环，对所述统计环进行迭代处理，当统计环与原始梯度图相交则基于统计环的内径对所述原始特征图像进行截图处理，得到待处理图像，并确定所述待处理图像的待处理图像参数：
获取与所述目标特征相对应的标准图像，确定所述标准图像对应的标准区域，并获取所述标准区域对应的标准图像参数；
根据待处理参数和所述标准图像参数，对所述待处理图像进行迁移处理，获取迁移图像；
对所述迁移图像进行限制对比度自适应直方图均衡处理，获取目标增强图像。
一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，所述计算机程序被处理器执行时实现如下步骤：
获取初始图像，对所述初始图像进行预处理，获取包含目标特征的原始特征图像；
采用边缘检测算法对所述原始特征图像进行边缘检测，获取原始梯度图，基于所述原始特征图获取统计环，对所述统计环进行迭代处理，当统计环与原始梯度图相交则基于统计环的内径对所述原始特征图像进行截图处理，得到待处理图像，并确定所述待处理图像的待处理图像参数：
获取与所述目标特征相对应的标准图像，确定所述标准图像对应的标准区域，并获取所述标准区域对应的标准图像参数；
根据待处理参数和所述标准图像参数，对所述待处理图像进行迁移处理，获取迁移图像；
对所述迁移图像进行限制对比度自适应直方图均衡处理，获取目标增强图像。
附图说明
为了更清楚地说明本申请实施例的技术方案，下面将对本申请实施例的描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
图1是本申请一实施例中基于人工智能的图像增强处理方法的一应用环境示意图；
图2是本申请一实施例中基于人工智能的图像增强处理方法的一流程图；
图3是本申请一实施例中基于人工智能的图像增强处理方法的另一流程图；
图4是本申请一实施例中基于人工智能的图像增强处理方法的另一流程图；
图5是本申请一实施例中基于人工智能的图像增强处理方法的另一流程图；
图6是本申请一实施例中基于人工智能的图像增强处理方法的另一流程图；
图7是本申请一实施例中基于人工智能的图像增强处理方法的另一流程图；
图8是本申请一实施例中基于人工智能的图像增强处理装置的一原理框图；
图9是本申请一实施例中计算机设备的一示意图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
本申请实施例提供的基于人工智能的图像增强处理方法，该基于人工智能的图像增强处理方法可应用如图1所示的应用环境中。具体地，该基于人工智能的图像增强处理方法应用在基于人工智能的图像增强处理系统中，该基于人工智能的图像增强处理系统包括如 图1所示的客户端和服务器，客户端与服务器通过网络进行通信，用于实现精准分割目标特征且图像质量较高的图像。其中，客户端又称为用户端，是指与服务器相对应,为客户提供本地服务的程序。客户端可安装在但不限于各种个人计算机、笔记本电脑、智能手机、平板电脑和便携式可穿戴设备上。服务器可以用独立的服务器或者是多个服务器组成的服务器集群来实现。
在一实施例中，如图2所示，提供一种基于人工智能的图像增强处理方法，以该方法应用在图1中的服务器为例进行说明，包括如下步骤：
S201：获取初始图像，对初始图像进行预处理，获取包含目标特征的原始特征图像。
其中，目标特征是指需要采用计算机视觉技术进行识别和分析的人体组织，以便后续确定用户的症状，目标特征可以是眼睛和口腔等。
初始图像是采用图像拍摄设备拍摄的人脸五官的图像。该图像拍摄设备可以是手机和医疗图像拍摄设备等。
原始特征图像是指以目标特征的长度和宽度作为裁剪尺寸对初始图像进行裁剪处理，得到以目标特征为中心的矩形图像，可以理解地，原始特征图像中包含非目标特征，即原始特征图像中包含噪声。本实施例中，截取初始图像的目标特征，以便后续针对目标特征进行处理，减少后续的图像处理的处理区域，减少噪声干扰。
S202：采用边缘检测算法对原始特征图像进行边缘检测，获取原始梯度图，基于原始特征图获取统计环，对统计环进行迭代处理，当统计环与原始梯度图相交则基于统计环的内径对原始特征图像进行截图处理，得到待处理图像，并确定待处理图像的待处理图像参数。
其中，边缘检测算法是用于检测图像上灰度变化最明显的区域的算法，该边缘检测算法包括但不限于Sobel算法、Laplace边缘检测算法和Canny边缘检测算法。
原始梯度图是表示原始特征图像的目标特征边缘的图像。
统计环是内径大于原始梯度图的环，也就是说，原始梯度图位于统计环内。迭代梯度值是指统计环的环内区域(即统计环外径与统计环内径的差值对应的区域)的梯度值。具体地，将统计环的环内区域打上掩码1，采用统计环的环外区域打上掩码0，将原始梯度图与统计环进行乘积运算，并计算统计环的环内区域的掩码和得到迭代梯度值。需要说明地是，统计环的中心与原始梯度图的中心重合。
待处理图像是仅包含目标特征的图像。本实施例中采用边缘检测算法对原始特征图进行处理，从而得到原始梯度图，然后根据统计环和原始梯度图确定目标特征的精准的位置，以获取仅包含目标特征的待处理图像，以便根据计算机视觉技术可以对精准地待处理图像进行识别和分析，确保识别和分析结果更加精确。
待处理图像参数是待处理图像对应的像素值，该待处理图像参数可以是待处理像素平均值和待处理像素方差等。
具体地，由于采用边缘检测算法的原始梯度图中，边缘较为模糊，边缘可能是连续的、也可能是不连续的或者与原始梯度图的中心距离不一致的，若直接根据原始梯度图对原始特征图进行分割得到目标特征，则得到的目标特征往往是不精准，存在不完整的或者非目标特征过多的问题，对于采用计算机视觉技术对目标特征进行识别，得到的识别结果存在误差。本实施例中，采用边缘检测算法对原始特征图像进行检测，以得到目标特征的边缘，即得到原始梯度图，对统计环和原始梯度图进行乘积运算，并对统计环的内径进行缩小迭代处理，计算缩小迭代处理过程中统计环的环内区域的掩码和，以形成迭代梯度值，由于原始梯度图具有边缘梯度值的，因此，若统计环的内径没有与原始梯度图相交，则迭代梯度值为0，若统计环的内径与原始梯度图相交，则迭代梯度值突增。当迭代梯度值突增时，则表示统计环与原始梯度图相交，此时以原始特征图的中心为圆心，以统计环内径的一半在原始特征图上画圆，并截图，得到仅包含目标特征的待处理图像，实现精准地分割出目 标特征，以便根据计算机视觉技术可以对精准地待处理图像进行识别和分析，确保识别和分析结果更加精确。接着计算待处理图像的待处理图像参数，以便后续对待处理图像进行迁移处理，得到图像质量较佳的图像，以提升图像处理的效果。
S203：获取与目标特征相对应的标准图像，确定标准图像对应的标准区域，并获取标准区域对应的标准图像参数。
其中，标准图像是指亮度、对比度和颜色空间质量高的图像，采用计算机视觉技术对标准图像进行识别可以得到较佳的识别效果。
标准图像参数是标准图像中目标特征对应的像素参数。该标准图像参数包括标准像素平均值和标准像素方差。
本实施例中，获取标准区域对应的标准图像参数，以便根据标准图像参数对待处理图像进行迁移处理，到便于识别的迁移图像。
S204：根据待处理参数和标准图像参数，对待处理图像进行迁移处理，获取迁移图像。
其中，迁移图像是指对待处理图像进行迁移处理得到的图像，该迁移图像具有较佳的亮度和对比度。
具体地，获取待处理图像上的任一点的待处理像素，以便根据待处理像素、待处理参数和标准图像参数进行计算得到待处理像素对应的迁移像素，根据迁移像素形成迁移图像。本实施例中，对待处理图像进行迁移处理，以使待处理图像的亮度和对比度与标准图像的亮度和对比度一致，使得待处理图像规范化。
S205：对迁移图像进行限制对比度自适应直方图均衡处理，获取目标增强图像。
其中，限制对比度自适应直方图均衡是提升图像的对比度的一种计算机图像处理技术，可以提升迁移图像的局部细节对比度，使得图像更加容易被识别，以提升基于人工智能的图像增强处理的效果。
具体地，对计算迁移图像进行分块处理，以得到多个待处理分块图，计算每一待处理分块图的分块直方图，并对每个分块直方图中幅值高于预设阈值的部分进行限幅，得到裁剪直方图，对裁剪直方图进行直方图均衡以得到目标增强图像，确保识别效果较佳。
本实施例所提供的基于人工智能的图像增强处理方法，获取初始图像，对初始图像进行预处理，获取包含目标特征的原始特征图像，以便后续针对目标特征进行处理，减少后续的图像处理的处理区域，减少噪声干扰。采用边缘检测算法对原始特征图像进行边缘检测，获取原始梯度图，基于原始特征图获取统计环，对统计环进行迭代处理，当统计环与原始梯度图相交则基于统计环的内径对原始特征图像进行截图处理，得到待处理图像，实现精准地分割出目标特征，以便根据计算机视觉技术可以对精准地待处理图像进行识别和分析，确保识别和分析结果更加精确；并确定待处理图像的待处理图像参数，以便后续对待处理图像进行迁移处理，得到图像质量较佳的图像，以提升图像处理的效果。获取与目标特征相对应的标准图像，确定标准图像对应的标准区域，并获取标准区域对应的标准图像参数，以便根据标准图像参数对待处理图像进行迁移处理，到便于识别的迁移图像。根据待处理参数和标准图像参数，对待处理图像进行迁移处理，获取迁移图像，以使待处理图像的亮度和对比度与标准图像的亮度和对比度一致，使得待处理图像规范化。对迁移图像进行限制对比度自适应直方图均衡处理，获取目标增强图像，确保识别效果较佳。
在一实施例中，步骤S202，即采用边缘检测算法对原始特征图像进行边缘检测，获取原始梯度图，基于原始特征图获取统计环，对统计环进行迭代处理，当统计环与原始梯度图相交则基于统计环的内径对原始特征图像进行截图处理，得到待处理图像，包括：
S301：采用Sobel算法计算原始特征图像中水平方向的水平梯度和垂直方向的垂直梯度。
其中，Sobel算法是一种离散性差分算法，它结合了高斯平滑和微分求导，用来运算图像亮度函数的灰度近似值。在图像的任何一点使用Sobel算子，将会产生对应的灰度矢 量或是其法矢量。
本实施例中，采用Sobel算法得到原始特征图像的水平梯度和垂直梯度，以便根据水平梯度和垂直梯度得到原始特征图像中目标特征边缘的大概位置。
S302：对水平梯度和垂直梯度进行加权处理，获取原始梯度图。
本实施例对水平梯度和垂直梯度进行以便根据水平梯度和垂直梯度进行加权处理，得到原始梯度图，从而得到目标特征边缘的大概位置，以便后续根据原始梯度图得到目标特征的精确位置。需要说明地是，Sobel算法得到的原始梯度图的边缘边缘较为模糊，可能是连续的、也可能是不连续的或者与原始梯度图的中心距离不一致的。
S303：根据原始特征图像的长边长度和预设间距，获取统计环。
其中，预设间距是预设的间距，用于根据原始特征图像长边长度(长边是指原始特征图中较长的一边)和预设间距确定统计环。具体地，以原始特征图中心为统计环的中心，以原始特征图长边长度的作为统计环的外径，以外径与预设间距的差值作为统计环的内径，以得到统计环。
本实施例中，利用统计环可以获取原始特征图像中目标特征的精准位置，以便精准分割出待处理图像。
S304：对统计环的内径做缩小迭代处理，获取统计环对应的迭代梯度值。
具体地，对统计环的内径做缩小迭代处理，在每一缩小处理时需要统计统计环的环内迭代梯度值，以便判断统计环是否与原始梯度图相交，以精准地从原始特征图中分割出目标特征的精准位置。
S305：若迭代梯度值大于预设梯度值，则统计环的内径与原始梯度图相交，将统计环的内圆确定为待处理图像。
本实施例中，当迭代梯度值大于预设梯度值时，则统计环的内径与原始梯度图相交，因此，以原始梯度的中心为圆形，以统计环的内径为直径，并截图，以得到待处理图像，从而实现精准地确定目标特征的位置，确保后续的图像处理具有针对性，减少干扰。
本实施例所提供的基于人工智能的图像增强处理方法，采用Sobel算法计算原始特征图像中水平方向的水平梯度和垂直方向的垂直梯度，对水平梯度和垂直梯度进行加权处理，获取原始梯度图，得到原始特征图像中目标特征边缘的大概位置。根据原始特征图像的长边长度和预设间距，获取统计环，对统计环的内径做缩小迭代处理，获取统计环对应的迭代梯度值，以便判断统计环是否与原始梯度图相交，以精准地从原始特征图中分割出目标特征的精准位置。若迭代梯度值大于预设梯度值，则统计环的内径与原始梯度图相交，将统计环的内圆确定为待处理图像，从而实现精准地确定目标特征的位置，确保后续的图像处理具有针对性，减少干扰。
在一实施例中，如图4所示，步骤S204，即根据待处理参数和标准图像参数，对待处理图像进行迁移处理，获取迁移图像，包括：
S401：获取待处理图像的每一待处理像素。
其中，待处理像素是指待处理图像中的像素。
S402：采用迁移公式对待处理像素、待处理参数和标准图像参数进行处理，获取待处理像素对应的迁移像素。
其中，迁移像素是待处理像素经过处理后的像素，一个迁移像素与一个待处理像素相对应，以形成与待处理图像相对应的迁移图像。
具体地，获取待处理图像中的每一待处理像素，将待处理像素、待处理参数和标准图像参数代入迁移公式中，得到与待处理像素对应的迁移像素，为后续形成迁移图像提供技术支持。
S403：基于迁移像素，形成迁移图像。
本实施例中，根据迁移像素形成迁移图像，提升迁移图像的亮度和保证颜色空间更佳， 减少干扰，确保后续采用计算机视觉技术进行图像识别的识别效果更好。
本实施例所提供的基于人工智能的图像增强处理方法，获取待处理图像的每一待处理像素。采用迁移公式对待处理像素、待处理参数和标准图像参数进行处理，获取待处理像素对应的迁移像素，为后续形成迁移图像提供技术支持。基于迁移像素，形成迁移图像，提升迁移图像的亮度和保证颜色空间更佳，减少干扰，确保后续采用计算机视觉技术进行图像识别的识别效果更好。
在一实施例中，待处理图像参数包括待处理像素平均值和待处理像素方差。标准图像参数包括标准像素平均值和标准像素方差。
步骤S402，即采用迁移公式对待处理像素、待处理参数和标准图像参数进行处理，获取待处理像素对应的迁移像素，包括：
将待处理像素、待处理像素平均值、待处理像素方差、标准像素平均值和标准像素方差代入迁移公式，获取与待处理像素对应的迁移像素。
其中，待处理像素平均值是指待处理图像中所有像素的平均值。待处理像素方差是指待处理图像中所有像素的方差。标准像素平均值是指标准区域中所有像素的平均值。标准像素方差是指标准区域中所有像素的方差。迁移公式为：
img output＝(img input-mean input)/(std input*std template)+mean template
其中，img output为迁移像素，img input为待处理像素，mean input为待处理像素平均值，std input为待处理像素方差；mean template为标准像素平均值，std template为标准像素方差。
本实施例中，利用迁移公式，实现对待处理图像进行迁移处理，形成迁移图像，以确保待迁移图像的亮度和颜色空间较佳，确保采用计算机视觉技术的识别效果较佳。
在一实施例中，如图5所示，步骤S205，即对迁移图像进行限制对比度自适应直方图均衡处理，获取目标增强图像，包括：
S501：基于迁移图像获取对应的通道图像。
其中，通道图像是指L通道对应的图像，该通道图像包含亮度信息，以便根据亮度信息进行处理，提升迁移图像的局部细节对比信息，确保得到的目标增强图像具有更高的清晰度，使得图像增强处理效果更佳，因此，后续计算机视觉技术识别效果更佳。
具体地，将迁移图像转化为Lab通道，仅对L通道进行处理，以便根据亮度信息进行处理，提升迁移图像的局部细节对比信息，避免采用RGB三通道分开处理导致严重的偏色。
S502：对通道图像进行分块处理，获取待处理分块图。
具体地，将通道图像进行8*8均匀分块，得到64个待处理分块图，以便针对每个待处理分块图分别处理，提升待处理分块图的局部细节对比度信息，避免对待处理分块图进行处理导致没有充分利用整个灰度动态范围，而不能有效地提高局部对比度的问题。
S503：采用直方图函数对待处理分块图进行处理，获取待处理分块图对应的分块直方图。
其中，直方图函数是将待处理分块图转化为分块直方图的函数。
本实施例中，通过直方图函数可以快速地得到分块直方图，以便根据分块直方图的像素分布情况进行灰度变换，达到提高图像对比度的目的。
S504：基于预设阈值对分块直方图裁剪，获取裁剪直方图，对裁剪直方图进行均衡化，获取目标增强图像。
其中，预设阈值是用于分块直方图进行裁剪的值。
具体地，根据分块直方图裁剪得到映射函数，在待处理分块图的每个像素点的值由它周围4个分块直方图的映射函数的映射函数值进行双线性插值得到，以根据映射函数值对待处理分块图的像素进行变换，提升局部细节对比度信息，从而得到目标增强图像。
在一实施例中，如图6所示，步骤S504，即基于预设阈值对分块直方图裁剪，获取 裁剪直方图，包括：
S601：获取待处理分块图对应的灰度值。
S602：将高于预设阈值的灰度值平均分配给所有分块直方图，获取裁剪直方图。
具体地，由于部分灰度值太高，若直接根据分块直方图得到映射曲线，导致映射曲线斜率过高，将所有灰度值都映射到整个灰度轴的右侧，导致图像失真，因此，需要将高于预设阈值的灰度值平均分配给所有分块直方图，以限制局部对比度，保证后续均衡化得到目标增强图像的局部细节对比度信息清晰。
在一实施例中，如图7所示，在步骤S201之前，即在获取包含目标特征的原始特征图像之前，基于人工智能的图像增强处理方法还包括：
S701：获取验证图像和与验证图像相对应的至少两个待选图像。
其中，待选图像是指预先采集的高质量模板图像，对待选图像进行筛选以确定质量最好的图像作为标准图像。
验证图像是用于对验证待选图像的质量进行验证的图像。
S702：执行步骤S201-S205对验证图像进行处理，获取验证图像对应的待测图像。
待测图像是依据待选图像对验证图像进行增强处理后获得的图像，用于验证待选图像的质量。
例如，验证图像为a1、a2和a3；待选图像为b1、b2、b3和b4；根据待选图像b1对验证图像a1、a2和a3进行迁移处理；根据待选图像b2对验证图像a1、a2和a3进行迁移处理；根据待选图像b3对验证图像a1、a2和a3进行迁移处理；根据待选图像b4对验证图像a1、a2和a3进行迁移处理，则得到迁移图像，对迁移图像进行限制对比度自适应直方图均衡处理，获取与待选图像对应b1的待测图像c11、c21和c31；与待选图像b2对应的待测图像c12、c22和c32；与待选图像b3对应的待测图像c13、c23和c33；与待选图像b4对应的待测图像c14、c24和c34。
S703：将待测图像输入基于待测图像对应的待选图像生成的MaskRCNN模型进行检测，获取检测结果。
其中，检测结果是采用MaskRCNN模型对待选图像进行扫描，并生成待选图像中目标特征的边界框和掩码的结果，以对待选图像进行筛选。该检测结果表示MaskRCNN模型在待选图像中所框选出来的部分为目标特征的比例。例如，该检测结果可以为100％，即MaskRCNN模型可以准确地检测出待选图像中的目标特征，即检测结果为准确；或者该检测结果可以为50％，即MaskRCNN模型可以检测出待选图像中的目标特征的50％，即检测结果为错误。
例如，将c11、c21和c31输入MaskRCNN模型1中，将c11、c21和c31输入MaskRCNN模型2中，将c11、c21和c31输入MaskRCNN模型3中，将c11、c21和c31输入MaskRCNN模型4中，以得到每一待测图像的检测效果。
S704：根据检测结果，确定每一待选图像的检测精度，将检测精度最高的待选图像确定为标准图像。
其中，检测精度是用于表示检测结果是否准确的精度，即该检测精度为检测结果为准确的数量除以总的检测结果数量。
具体地，统计根据每一待选图像进行处理的待测图像对应的检测精度，选取检测精度最高的待选图像确定为标准图像，以便待处理图像提供质量最高的参照图像。
本实施例中，获取验证图像和与验证图像相对应的至少两个待选图像；执行步骤S201-S205对验证图像进行处理，获取验证图像对应的待测图像，以验证待选图像的质量。将待测图像输入基于待测图像对应的待选图像生成的MaskRCNN模型进行检测，获取检测结果，以对待选图像进行筛选。根据检测结果，确定每一待选图像的检测精度，将检测精度最高的待选图像确定为标准图像，以便待处理图像提供质量最高的参照图像。
应理解，上述实施例中各步骤的序号的大小并不意味着执行顺序的先后，各过程的执行顺序应以其功能和内在逻辑确定，而不应对本申请实施例的实施过程构成任何限定。
在一实施例中，提供一种基于人工智能的图像增强处理装置，该基于人工智能的图像增强处理装置与上述实施例中基于人工智能的图像增强处理方法一一对应。如图8所示，该基于人工智能的图像增强处理装置包括原始特征图像获取模块801、待处理图像获取模块802、标准图像参数获取模块803、迁移图像获取模块804和目标增强图像获取模块805。各功能模块详细说明如下：
原始特征图像获取模块801，用于获取初始图像，对初始图像进行预处理，获取包含目标特征的原始特征图像。
待处理图像获取模块802，用于采用边缘检测算法对原始特征图像进行边缘检测，获取原始梯度图，基于原始特征图获取统计环，对统计环进行迭代处理，当统计环与原始梯度图相交则基于统计环的内径对原始特征图像进行截图处理，得到待处理图像，并确定待处理图像的待处理图像参数：
标准图像参数获取模块803，用于获取与目标特征相对应的标准图像，确定标准图像对应的标准区域，并获取标准区域对应的标准图像参数。
迁移图像获取模块804，用于根据待处理参数和标准图像参数，对待处理图像进行迁移处理，获取迁移图像。
目标增强图像获取模块805，用于对迁移图像进行限制对比度自适应直方图均衡处理，获取目标增强图像。
优选地，待处理图像获取模块802，包括：梯度计算单元、原始梯度图获取单元、统计环获取单元、迭代梯度值获取单元和待处理图像获取单元。
梯度计算单元，用于采用Sobel算法计算原始特征图像中水平方向的水平梯度和垂直方向的垂直梯度。
原始梯度图获取单元，用于对水平梯度和垂直梯度进行加权处理，获取原始梯度图。
统计环获取单元，用于根据原始特征图像的长边长度和预设间距，获取统计环。
迭代梯度值获取单元，用于对统计环的内径做缩小迭代处理，获取统计环对应的迭代梯度值。
待处理图像获取单元，用于若迭代梯度值大于预设梯度值，则统计环的内径与原始梯度图相交，将统计环的内圆确定为待处理图像。
优选地，迁移图像获取模块804，包括：待处理像素获取单元、迁移像素获取单元和迁移图像获取单元。
待处理像素获取单元，用于获取待处理图像的每一待处理像素。
迁移像素获取单元，用于采用迁移公式对待处理像素、待处理参数和标准图像参数进行处理，获取待处理像素对应的迁移像素。
迁移图像获取单元，用于基于迁移像素，形成迁移图像。
优选地，待处理图像参数包括待处理像素平均值和待处理像素方差。标准图像参数包括标准像素平均值和标准像素方差。
迁移像素获取单元，包括：迁移像素计算单元。
将待处理像素、待处理像素平均值、待处理像素方差、标准像素平均值和标准像素方差代入迁移公式，获取与待处理像素对应的迁移像素。
优选地，目标增强图像获取模块805，包括：通道图像获取单元、待处理分块图获取单元、分块直方图获取单元和均衡化处理单元。
通道图像获取单元，用于基于迁移图像获取对应的通道图像。
待处理分块图获取单元，用于对通道图像进行分块处理，获取待处理分块图。
分块直方图获取单元，用于采用直方图函数对待处理分块图进行处理，获取待处理分 块图对应的分块直方图。
均衡化处理单元，用于基于预设阈值对分块直方图裁剪，获取裁剪直方图，对裁剪直方图进行均衡化，获取目标增强图像。
优选地，均衡化处理单元，包括：
灰度值获取单元，用于获取待处理分块图对应的灰度值。
裁剪直方图获取单元，用于将高于预设阈值的灰度值平均分配给所有分块直方图，获取裁剪直方图。
优选地，在原始特征图像获取模块801之前，基于人工智能的图像增强处理方法装置还包括：待选图像获取模块、待测图像获取模块、检测结果获取模块和标准图像获取模块。
待选图像获取模块，用于获取验证图像和与验证图像相对应的至少两个待选图像。
待测图像获取模块，用于执行如权利要求1的步骤对验证图像进行处理，获取验证图像对应的待测图像。
检测结果获取模块，用于将待测图像输入基于待测图像对应的待选图像生成的MaskRCNN模型进行检测，获取检测结果。
标准图像获取模块，用于根据检测结果，确定每一待选图像的检测精度，将检测精度最高的待选图像确定为标准图像。
关于基于人工智能的图像增强处理装置的具体限定可以参见上文中对于基于人工智能的图像增强处理方法的限定，在此不再赘述。上述基于人工智能的图像增强处理装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。
在一个实施例中，提供了一种计算机设备，该计算机设备可以是服务器，其内部结构图可以如图9所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中，该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性或易失性存储介质、内存储器。该非易失性或易失性存储介质存储有操作系统、计算机程序和数据库。该内存储器为非易失性或易失性存储介质中的操作系统和计算机程序的运行提供环境。该计算机设备的数据库用于存储目标增强图像。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机程序被处理器执行时以实现一种基于人工智能的图像增强处理方法。
在一个实施例中，提供了一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，处理器执行计算机程序时实现上述实施例中基于人工智能的图像增强处理方法的步骤，例如图2所示的步骤S201-S205，或者图3至图7中所示的步骤，为避免重复，这里不再赘述。或者，处理器执行计算机程序时实现基于人工智能的图像增强处理装置这一实施例中的各模块/单元的功能，例如图8所示的原始特征图像获取模块801、待处理图像获取模块802、标准图像参数获取模块803、迁移图像获取模块804和目标增强图像获取模块80的功能，为避免重复，这里不再赘述。
在一实施例中，提供一计算机可读存储介质，所述计算机可读存储介质可以是易失性，也可以是非易失性，该计算机可读存储介质上存储有计算机程序，该计算机程序被处理器执行时实现上述实施例中基于人工智能的图像增强处理方法的步骤，例如图2所示的步骤S201-S205，或者图3至图7中所示的步骤，为避免重复，这里不再赘述。或者，处理器执行计算机程序时实现基于人工智能的图像增强处理装置这一实施例中的各模块/单元的功能，例如图8所示的原始特征图像获取模块801、待处理图像获取模块802、标准图像参数获取模块803、迁移图像获取模块804和目标增强图像获取模块80的功能，为避免重复，这里不再赘述。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过 计算机程序来指令相关的硬件来完成，所述的计算机程序可存储于一非易失性计算机可读取存储介质中，该计算机程序在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，仅以上述各功能单元、模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能单元、模块完成，即将所述装置的内部结构划分成不同的功能单元或模块，以完成以上描述的全部或者部分功能。
以上所述实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本申请各实施例技术方案的精神和范围，均应包含在本申请的保护范围之内。

Claims (20)

  1. 一种基于人工智能的图像增强处理方法，其中，包括：
    获取初始图像，对所述初始图像进行预处理，获取包含目标特征的原始特征图像；
    采用边缘检测算法对所述原始特征图像进行边缘检测，获取原始梯度图，基于所述原始特征图获取统计环，对所述统计环进行迭代处理，当统计环与原始梯度图相交则基于统计环的内径对所述原始特征图像进行截图处理，得到待处理图像，并确定所述待处理图像的待处理图像参数：
    获取与所述目标特征相对应的标准图像，确定所述标准图像对应的标准区域，并获取所述标准区域对应的标准图像参数；
    根据待处理参数和所述标准图像参数，对所述待处理图像进行迁移处理，获取迁移图像；
    对所述迁移图像进行限制对比度自适应直方图均衡处理，获取目标增强图像。
  2. 如权利要求1所述的基于人工智能的图像增强处理方法，其中，所述采用边缘检测算法对所述原始特征图像进行边缘检测，获取原始梯度图，基于所述原始特征图获取统计环，对所述统计环进行迭代处理，当统计环与原始梯度图相交则基于统计环的内径对所述原始特征图像进行截图处理，得到待处理图像，并确定所述待处理图像的待处理图像参数，包括：
    采用Sobel算法计算所述原始特征图像中水平方向的水平梯度和垂直方向的垂直梯度；
    对所述水平梯度和所述垂直梯度进行加权处理，获取原始梯度图；
    根据原始特征图像的长边长度和预设间距，获取统计环；
    对所述统计环的内径做缩小迭代处理，获取所述统计环对应的迭代梯度值；
    若所述迭代梯度值大于预设梯度值，则所述统计环的内径与所述原始梯度图相交，将所述统计环的内圆确定为待处理图像。
  3. 如权利要求1所述的基于人工智能的图像增强处理方法，其中，所述根据待处理参数和所述标准图像参数，对所述待处理图像进行迁移处理，获取迁移图像，包括：
    获取所述待处理图像的每一待处理像素；
    采用迁移公式对所述待处理像素、所述待处理参数和所述标准图像参数进行处理，获取所述待处理像素对应的迁移像素；
    基于所述迁移像素，形成迁移图像。
  4. 如权利要求3所述的基于人工智能的图像增强处理方法，其中，待处理图像参数包括待处理像素平均值和待处理像素方差；标准图像参数包括标准像素平均值和所述标准像素方差；
    所述采用迁移公式对所述待处理像素、所述待处理参数和所述标准图像参数进行处理，获取所述待处理像素对应的迁移像素，包括：
    将所述待处理像素、所述待处理像素平均值、所述待处理像素方差、所述标准像素平均值和所述标准像素方差代入迁移公式，获取与所述待处理像素对应的迁移像素。
  5. 如权利要求1所述的基于人工智能的图像增强处理方法，其中，所述对所述迁移图像进行限制对比度自适应直方图均衡处理，获取目标增强图像，包括：
    基于所述迁移图像获取对应的通道图像；
    对所述通道图像进行分块处理，获取待处理分块图；
    采用直方图函数对所述待处理分块图进行处理，获取待处理分块图对应的分块直方图；
    基于预设阈值对所述分块直方图裁剪，获取裁剪直方图，对所述裁剪直方图进行均衡化，获取目标增强图像。
  6. 如权利要求5所述的基于人工智能的图像增强处理方法，其中，所述基于预设阈值 对所述分块直方图裁剪，获取裁剪直方图，包括：
    获取所述待处理分块图对应的灰度值；
    将高于所述预设阈值的灰度值平均分配给所有所述分块直方图，获取裁剪直方图。
  7. 如权利要求1所述的基于人工智能的图像增强处理方法，其中，在所述获取包含目标特征的原始特征图像之前，所述基于人工智能的图像增强处理方法还包括：
    获取验证图像和与所述验证图像相对应的至少两个待选图像；
    执行如权利要求1的步骤对所述验证图像进行处理，获取所述验证图像对应的待测图像；
    将所述待测图像输入基于所述待测图像对应的待选图像生成的MaskRCNN模型进行检测，获取检测结果；
    根据所述检测结果，确定每一所述待选图像的检测精度，将检测精度最高的所述待选图像确定为标准图像。
  8. 一种基于人工智能的图像增强处理装置，其中，包括：
    原始特征图像获取模块，用于获取初始图像，对所述初始图像进行预处理，获取包含目标特征的原始特征图像；
    待处理图像获取模块，用于采用边缘检测算法对所述原始特征图像进行边缘检测，获取原始梯度图，基于所述原始特征图获取统计环，对所述统计环进行迭代处理，当统计环与原始梯度图相交则基于统计环的内径对所述原始特征图像进行截图处理，得到待处理图像，并确定所述待处理图像的待处理图像参数：
    标准图像参数获取模块，用于获取与所述目标特征相对应的标准图像，确定所述标准图像对应的标准区域，并获取所述标准区域对应的标准图像参数。
    迁移图像获取模块，用于根据待处理参数和所述标准图像参数，对所述待处理图像进行迁移处理，获取迁移图像；
    目标增强图像获取模块，用于对所述迁移图像进行限制对比度自适应直方图均衡处理，获取目标增强图像。
  9. 一种计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序，其中，所述处理器执行所述计算机程序时实现如下步骤：
    获取初始图像，对所述初始图像进行预处理，获取包含目标特征的原始特征图像；
    采用边缘检测算法对所述原始特征图像进行边缘检测，获取原始梯度图，基于所述原始特征图获取统计环，对所述统计环进行迭代处理，当统计环与原始梯度图相交则基于统计环的内径对所述原始特征图像进行截图处理，得到待处理图像，并确定所述待处理图像的待处理图像参数：
    获取与所述目标特征相对应的标准图像，确定所述标准图像对应的标准区域，并获取所述标准区域对应的标准图像参数；
    根据待处理参数和所述标准图像参数，对所述待处理图像进行迁移处理，获取迁移图像；
    对所述迁移图像进行限制对比度自适应直方图均衡处理，获取目标增强图像。
  10. 如权利要求9所述的计算机设备，其中，所述采用边缘检测算法对所述原始特征图像进行边缘检测，获取原始梯度图，基于所述原始特征图获取统计环，对所述统计环进行迭代处理，当统计环与原始梯度图相交则基于统计环的内径对所述原始特征图像进行截图处理，得到待处理图像，并确定所述待处理图像的待处理图像参数，包括：
    采用Sobel算法计算所述原始特征图像中水平方向的水平梯度和垂直方向的垂直梯度；
    对所述水平梯度和所述垂直梯度进行加权处理，获取原始梯度图；
    根据原始特征图像的长边长度和预设间距，获取统计环；
    对所述统计环的内径做缩小迭代处理，获取所述统计环对应的迭代梯度值；
    若所述迭代梯度值大于预设梯度值，则所述统计环的内径与所述原始梯度图相交，将所述统计环的内圆确定为待处理图像。
  11. 如权利要求9所述的计算机设备，其中，所述根据待处理参数和所述标准图像参数，对所述待处理图像进行迁移处理，获取迁移图像，包括：
    获取所述待处理图像的每一待处理像素；
    采用迁移公式对所述待处理像素、所述待处理参数和所述标准图像参数进行处理，获取所述待处理像素对应的迁移像素；
    基于所述迁移像素，形成迁移图像。
  12. 如权利要求11所述的计算机设备，其中，待处理图像参数包括待处理像素平均值和待处理像素方差；标准图像参数包括标准像素平均值和所述标准像素方差；
    所述采用迁移公式对所述待处理像素、所述待处理参数和所述标准图像参数进行处理，获取所述待处理像素对应的迁移像素，包括：
    将所述待处理像素、所述待处理像素平均值、所述待处理像素方差、所述标准像素平均值和所述标准像素方差代入迁移公式，获取与所述待处理像素对应的迁移像素。
  13. 如权利要求9所述的计算机设备，其中，所述对所述迁移图像进行限制对比度自适应直方图均衡处理，获取目标增强图像，包括：
    基于所述迁移图像获取对应的通道图像；
    对所述通道图像进行分块处理，获取待处理分块图；
    采用直方图函数对所述待处理分块图进行处理，获取待处理分块图对应的分块直方图；
    基于预设阈值对所述分块直方图裁剪，获取裁剪直方图，对所述裁剪直方图进行均衡化，获取目标增强图像。
  14. 如权利要求13所述的计算机设备，其中，所述基于预设阈值对所述分块直方图裁剪，获取裁剪直方图，包括：
    获取所述待处理分块图对应的灰度值；
    将高于所述预设阈值的灰度值平均分配给所有所述分块直方图，获取裁剪直方图。
  15. 如权利要求9所述的计算机设备，其中，在所述获取包含目标特征的原始特征图像之前，所述处理器执行所述计算机程序时还实现如下步骤：
    获取验证图像和与所述验证图像相对应的至少两个待选图像；
    执行如权利要求1的步骤对所述验证图像进行处理，获取所述验证图像对应的待测图像；
    将所述待测图像输入基于所述待测图像对应的待选图像生成的MaskRCNN模型进行检测，获取检测结果；
    根据所述检测结果，确定每一所述待选图像的检测精度，将检测精度最高的所述待选图像确定为标准图像。
  16. 一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，其中，所述计算机程序被处理器执行时实现如下步骤：
    获取初始图像，对所述初始图像进行预处理，获取包含目标特征的原始特征图像；
    采用边缘检测算法对所述原始特征图像进行边缘检测，获取原始梯度图，基于所述原始特征图获取统计环，对所述统计环进行迭代处理，当统计环与原始梯度图相交则基于统计环的内径对所述原始特征图像进行截图处理，得到待处理图像，并确定所述待处理图像的待处理图像参数：
    获取与所述目标特征相对应的标准图像，确定所述标准图像对应的标准区域，并获取所述标准区域对应的标准图像参数；
    根据待处理参数和所述标准图像参数，对所述待处理图像进行迁移处理，获取迁移图像；
    对所述迁移图像进行限制对比度自适应直方图均衡处理，获取目标增强图像。
  17. 如权利要求16所述的计算机可读存储介质，其中，所述采用边缘检测算法对所述原始特征图像进行边缘检测，获取原始梯度图，基于所述原始特征图获取统计环，对所述统计环进行迭代处理，当统计环与原始梯度图相交则基于统计环的内径对所述原始特征图像进行截图处理，得到待处理图像，并确定所述待处理图像的待处理图像参数，包括：
    采用Sobel算法计算所述原始特征图像中水平方向的水平梯度和垂直方向的垂直梯度；
    对所述水平梯度和所述垂直梯度进行加权处理，获取原始梯度图；
    根据原始特征图像的长边长度和预设间距，获取统计环；
    对所述统计环的内径做缩小迭代处理，获取所述统计环对应的迭代梯度值；
    若所述迭代梯度值大于预设梯度值，则所述统计环的内径与所述原始梯度图相交，将所述统计环的内圆确定为待处理图像。
  18. 如权利要求16所述的计算机可读存储介质，其中，所述根据待处理参数和所述标准图像参数，对所述待处理图像进行迁移处理，获取迁移图像，包括：
    获取所述待处理图像的每一待处理像素；
    采用迁移公式对所述待处理像素、所述待处理参数和所述标准图像参数进行处理，获取所述待处理像素对应的迁移像素；
    基于所述迁移像素，形成迁移图像。
  19. 如权利要求18所述的计算机可读存储介质，其中，待处理图像参数包括待处理像素平均值和待处理像素方差；标准图像参数包括标准像素平均值和所述标准像素方差；
    所述采用迁移公式对所述待处理像素、所述待处理参数和所述标准图像参数进行处理，获取所述待处理像素对应的迁移像素，包括：
    将所述待处理像素、所述待处理像素平均值、所述待处理像素方差、所述标准像素平均值和所述标准像素方差代入迁移公式，获取与所述待处理像素对应的迁移像素。
  20. 如权利要求16所述的计算机可读存储介质，其中，所述对所述迁移图像进行限制对比度自适应直方图均衡处理，获取目标增强图像，包括：
    基于所述迁移图像获取对应的通道图像；
    对所述通道图像进行分块处理，获取待处理分块图；
    采用直方图函数对所述待处理分块图进行处理，获取待处理分块图对应的分块直方图；
    基于预设阈值对所述分块直方图裁剪，获取裁剪直方图，对所述裁剪直方图进行均衡化，获取目标增强图像。
PCT/CN2020/112331 2020-07-31 2020-08-30 基于人工智能的图像增强处理方法、装置、设备及介质 WO2021189770A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/613,482 US20220245803A1 (en) 2020-07-31 2020-08-30 Image enhancement processing method, device, equipment, and medium based on artificial intelligence

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010763141.4A CN111915541B (zh) 2020-07-31 2020-07-31 基于人工智能的图像增强处理方法、装置、设备及介质
CN202010763141.4 2020-07-31

Publications (1)

Publication Number Publication Date
WO2021189770A1 true WO2021189770A1 (zh) 2021-09-30

Family

ID=73287587

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/112331 WO2021189770A1 (zh) 2020-07-31 2020-08-30 基于人工智能的图像增强处理方法、装置、设备及介质

Country Status (3)

Country Link
US (1) US20220245803A1 (zh)
CN (1) CN111915541B (zh)
WO (1) WO2021189770A1 (zh)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114529459B (zh) * 2022-04-25 2022-08-02 东莞市兆丰精密仪器有限公司 一种对图像边缘进行增强处理的方法和系统及介质
CN115049661B (zh) * 2022-08-15 2022-11-25 深圳华声医疗技术股份有限公司 目标结构围度测量方法、装置、超声设备及存储介质
CN116012378B (zh) * 2023-03-24 2023-06-23 湖南东方钪业股份有限公司 一种用于增材制造的合金丝材的质量检测方法

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101763641A (zh) * 2009-12-29 2010-06-30 电子科技大学 一种模拟视觉机制的图像目标物体轮廓检测方法
CN107785061A (zh) * 2017-10-10 2018-03-09 东南大学 孤独症谱系障碍儿童情绪能力干预系统
CN107967442A (zh) * 2017-09-30 2018-04-27 广州智慧城市发展研究院 一种基于无监督学习和深层网络的指静脉识别方法及系统
CN108765347A (zh) * 2018-05-30 2018-11-06 长光卫星技术有限公司 一种适合遥感影像的色彩增强方法
CN110363782A (zh) * 2019-06-13 2019-10-22 平安科技（深圳）有限公司 一种基于边缘识别算法的区域识别方法、装置及电子设备
US20190340738A1 (en) * 2016-12-28 2019-11-07 Karl-Franzens-Universitat Graz Method and device for image processing
CN111199531A (zh) * 2019-12-27 2020-05-26 中国民航大学 基于泊松图像融合及图像风格化的交互式数据扩展方法

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101211459A (zh) * 2007-12-21 2008-07-02 上海广电集成电路有限公司 一种针对边界点进行处理的直方图均衡方法
CN101276461B (zh) * 2008-03-07 2011-11-02 北京航空航天大学 一种利用边缘特征的视频文本增强方法
KR101316196B1 (ko) * 2011-12-23 2013-10-08 연세대학교 산학협력단 입체 영상 화질 향상 장치 및 방법과 이에 관한 기록매체
CN103325098A (zh) * 2013-07-02 2013-09-25 南京理工大学 基于多尺度处理的高动态红外图像增强方法
CN108846319A (zh) * 2018-05-25 2018-11-20 平安科技（深圳）有限公司 基于直方图的虹膜图像增强方法、装置、设备及存储介质
US11055822B2 (en) * 2019-05-03 2021-07-06 International Business Machines Corporation Artificially intelligent, machine learning-based, image enhancement, processing, improvement and feedback algorithms

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101763641A (zh) * 2009-12-29 2010-06-30 电子科技大学 一种模拟视觉机制的图像目标物体轮廓检测方法
US20190340738A1 (en) * 2016-12-28 2019-11-07 Karl-Franzens-Universitat Graz Method and device for image processing
CN107967442A (zh) * 2017-09-30 2018-04-27 广州智慧城市发展研究院 一种基于无监督学习和深层网络的指静脉识别方法及系统
CN107785061A (zh) * 2017-10-10 2018-03-09 东南大学 孤独症谱系障碍儿童情绪能力干预系统
CN108765347A (zh) * 2018-05-30 2018-11-06 长光卫星技术有限公司 一种适合遥感影像的色彩增强方法
CN110363782A (zh) * 2019-06-13 2019-10-22 平安科技（深圳）有限公司 一种基于边缘识别算法的区域识别方法、装置及电子设备
CN111199531A (zh) * 2019-12-27 2020-05-26 中国民航大学 基于泊松图像融合及图像风格化的交互式数据扩展方法

Also Published As

Publication number Publication date
CN111915541B (zh) 2021-08-17
CN111915541A (zh) 2020-11-10
US20220245803A1 (en) 2022-08-04

Similar Documents

Publication Publication Date Title
WO2021017261A1 (zh) 识别模型训练方法、图像识别方法、装置、设备及介质
WO2021189770A1 (zh) 基于人工智能的图像增强处理方法、装置、设备及介质
CN110222787B (zh) 多尺度目标检测方法、装置、计算机设备及存储介质
CN109359575B (zh) 人脸检测方法、业务处理方法、装置、终端及介质
CN110826519B (zh) 人脸遮挡检测方法、装置、计算机设备及存储介质
WO2019223069A1 (zh) 基于直方图的虹膜图像增强方法、装置、设备及存储介质
WO2022199583A1 (zh) 图像处理方法、装置、计算机设备和存储介质
WO2021179471A1 (zh) 一种人脸模糊度检测方法、装置、计算机设备及存储介质
WO2021135500A1 (zh) 车损检测模型训练、车损检测方法、装置、设备及介质
US11501431B2 (en) Image processing method and apparatus and neural network model training method
WO2021175180A1 (zh) 视线确定方法、装置、电子设备和计算机可读存储介质
WO2021136368A1 (zh) 钼靶图像中胸大肌区域自动检测方法及装置
WO2019223068A1 (zh) 虹膜图像局部增强方法、装置、设备及存储介质
CN113781406B (zh) 电子元器件的划痕检测方法、装置及计算机设备
WO2021196721A1 (zh) 一种舱内环境的调整方法及装置
WO2020001219A1 (zh) 图像处理方法和装置、存储介质、电子设备
CN113469092B (zh) 字符识别模型生成方法、装置、计算机设备和存储介质
WO2022135574A1 (zh) 肤色检测方法、装置、移动终端和存储介质
CN107146231B (zh) 视网膜图像出血区域分割方法、装置和计算设备
JP2021174554A (ja) 画像深度確定方法及び生き物認識方法、回路、装置、記憶媒体
CN111145086A (zh) 一种图像处理方法、装置及电子设备
CN112884782A (zh) 生物对象分割方法、装置、计算机设备和存储介质
CN111444555B (zh) 一种测温信息显示方法、装置及终端设备
CN115131714A (zh) 视频图像智能检测分析方法及系统
JP6819445B2 (ja) 情報処理装置、制御方法、及びプログラム

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20927256

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20927256

Country of ref document: EP

Kind code of ref document: A1