CN118505825B - Cashmere product color measurement method and device based on image recognition - Google Patents

Abstract

The invention relates to the technical field of image processing, in particular to a cashmere product color measurement method and device based on image identification, wherein the method comprises the following steps: acquiring surface images of cashmere products, and acquiring images of each channel; performing super-pixel segmentation on each channel image, obtaining a plurality of super-pixel blocks of each channel image, and obtaining an average gray index of each super-pixel block of each channel image and a gray sequence of each pixel point of each channel image; according to the gray level sequence of each pixel point of each channel image and the average gray level index of each super pixel block of each channel image, the illumination influence degree of each pixel point of each channel image is obtained, the weight correction coefficient of each pixel point of each channel image is further obtained, the Gaussian filter kernel of each pixel point of each channel image is obtained in a self-adaptive mode according to the weight correction coefficient, and the Gaussian filter is carried out on each channel image.

Description

Cashmere product color measurement method and device based on image recognition

Technical Field

The present invention relates to the field of image processing technology, and more specifically, to a method and device for color measurement of cashmere products based on image recognition.

Background Art

Cashmere product color measurement is an important means to ensure product quality, control production costs, meet market demand and comply with quality standards. It is of great significance to the production and market competition of enterprises. The color of cashmere products is measured by image recognition technology. By collecting image data of cashmere products, color features are extracted and analyzed to achieve accurate measurement and evaluation of color. The color measurement of cashmere products can be performed more accurately, saving labor costs. However, since the surface of cashmere products is often not smooth enough, the grayscale of the surface will change complexly when affected by light, resulting in inaccurate color measurement results of cashmere products. Therefore, it is necessary to filter the surface image of cashmere products.

The patent application document with the publication number CN112651895A discloses an FPGA-based image Gaussian filtering method, which collects images through a camera; performs grayscale preprocessing on the collected images; processes the preprocessed images using a Gaussian filtering algorithm; and caches the processed images and displays them on a display.

In the process of using the above method to filter the cashmere product surface image to eliminate the influence of light, since the weight assigned by the Gaussian filter kernel to each pixel point of the cashmere product surface image is only related to the Euclidean distance between the pixels, the influence of light on the pixels is not taken into account, and the influence of light on each pixel point cannot be eliminated, which reduces the filtering effect of the cashmere product surface image. Therefore, the above method cannot be used directly to filter the cashmere product surface image.

Summary of the invention

In order to solve the problem that the weight assigned by the Gaussian filter kernel in the Gaussian filter algorithm to each pixel point of the cashmere product surface image is only related to the Euclidean distance between the pixel points, and the influence of light on the pixel points is not taken into account, resulting in a reduction in the filtering effect of the cashmere product surface image, the present invention proposes a cashmere product color measurement method and device based on image recognition.

In the first aspect, the present invention provides a cashmere product color measurement method based on image recognition, which adopts the following technical scheme:

Collecting a surface image of a cashmere product; acquiring an image of each channel according to the surface image of the cashmere product;

Perform superpixel segmentation on each channel image to obtain multiple superpixel blocks of each channel image; obtain the average grayscale index of each superpixel block of each channel image according to the difference between the average grayscale value of the pixels in each superpixel block of each channel image and the maximum value of the average grayscale value of the pixels in all superpixel blocks of each channel image; obtain the degree of illumination influence of each pixel point in each channel image:

,in, Represents the degree of illumination influence of the jth pixel of the nth channel image, Represents the average grayscale index of the superpixel block of the nth channel image where the jth pixel of the nth channel image is located; The number of pixels in the local window representing the jth pixel of the nth channel image; Represents the grayscale value of the vth pixel in the grayscale sequence of the jth pixel of the nth channel image; Represents the gray value of the v+1th pixel in the gray sequence of the jth pixel of the nth channel image;

Obtain a Gaussian filter kernel for each pixel of each channel image, wherein the weight of each pixel of the Gaussian filter kernel for each pixel of each channel image is proportional to the degree of illumination influence of each pixel of each channel image; perform Gaussian filtering on each channel image based on the Gaussian filter kernel for each pixel of each channel image to obtain a filtering result image of each channel.

Preferably, the step of performing superpixel segmentation on each channel image to obtain a plurality of superpixel blocks of each channel image includes:

Preset the superpixel block size m and obtain the size of the nth channel image, denoted as M N, then the number of superpixel blocks in the nth channel image , according to the number of super-pixel blocks of the n-th channel image, use the super-pixel segmentation algorithm to perform super-pixel segmentation on the n-th channel image to obtain several super-pixel blocks of the n-th channel image; Represents the ceiling function.

Each channel image is segmented into superpixels, and pixels with similar grayscale values are divided into a superpixel block. This makes it easier to measure whether each pixel in the superpixel block is affected by light by analyzing the average grayscale index of each superpixel block.

Preferably, the average grayscale index of each superpixel block of each channel image is obtained, and the calculation formula is:

;

In the formula, Represents the average grayscale index of the i-th superpixel block of the n-th channel image; Represents the average gray value of the pixels in the i-th superpixel block of the n-th channel image; Represents the maximum value of the average grayscale value of pixels in all superpixel blocks of the nth channel image; exp() represents an exponential function with a natural constant as the base.

The larger the average grayscale index value of each superpixel block in each channel image is, the more likely the superpixel block is a shadow area affected by illumination, which facilitates the subsequent acquisition of the illumination influence degree of each pixel point in each channel image based on this value.

Preferably, the acquisition of the local window includes:

The default local window size is , with each pixel of each channel image as the center pixel to construct a pixel with a size of The local window of is used as the local window of each pixel point of each channel image.

Preferably, the grayscale sequence is obtained by:

All the pixels in the local window of each pixel of each channel image are sorted in ascending order of grayscale value to obtain the grayscale sequence of each pixel of each channel image.

Preferably, the step of obtaining the Gaussian filter kernel of each pixel point of each channel image includes:

Get the weight correction coefficient of each pixel of each channel image;

The preset Gaussian filter window size is , obtain the weight of each pixel in the Gaussian filter kernel of each pixel of the n-th channel image according to the Gaussian function, and multiply the weight of each pixel in the Gaussian filter kernel of each pixel of the n-th channel image by the weight correction coefficient of each pixel of the n-th channel image, obtain the updated weight of each pixel in the Gaussian filter kernel of each pixel of the n-th channel image, and obtain the Gaussian filter kernel of each pixel of the n-th channel image.

After the weight of each pixel in the Gaussian filter kernel of each pixel of each channel image is corrected according to the weight correction coefficient of each pixel of each channel image, the Gaussian filter kernel of each pixel is adaptively used to make the subsequent filtering result more accurate.

Preferably, obtaining the weight correction coefficient of each pixel point of each channel image includes:

The illumination influence degree of each pixel point of each channel image is linearly normalized to obtain the weight correction coefficient of each pixel point of each channel image.

Preferably, the Gaussian filtering of each channel image by using a Gaussian filter kernel based on each pixel point of each channel image to obtain a filtering result image of each channel includes:

The number of filtering times d is preset, and the images of each channel are filtered using the Gaussian filtering algorithm according to the Gaussian filtering kernel of each pixel point of each channel image until the number of filtering times of each channel image reaches d, thereby obtaining the filtering result images of each channel.

The channel images are filtered according to the Gaussian filter kernel of each pixel point of each channel image to eliminate the influence of illumination, so that the filtering result of the image is more accurate.

In a second aspect, the present invention provides a cashmere product color measurement device based on image recognition, which adopts the following technical solution:

The cashmere product color measurement device based on image recognition comprises: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the cashmere product color measurement method based on image recognition is implemented.

By adopting the above technical solution, the above-mentioned cashmere product color measurement method based on image recognition is generated into a computer program and stored in a memory so as to be loaded and executed by a processor, thereby making a terminal device based on the memory and the processor for easy use.

The present invention has the following beneficial effects: the innovation of the present invention lies in obtaining the degree of illumination influence of each pixel point of each channel image, correcting the weight of each pixel point in the Gaussian filter kernel of each pixel point of each channel image, adaptively obtaining the Gaussian filter kernel of each pixel point of each channel image to filter each channel image, eliminating the influence of illumination, so that the filtering result of the image is more accurate, and the degree of illumination influence of each pixel point of each channel image is proportional to the complexity of grayscale distribution in the local range of each pixel point and the average grayscale index of the super pixel block where each pixel point is located. The larger the value of the illumination influence degree is, the more likely the pixel point is to be affected by illumination. The degree of correction of the weight of each pixel point in the Gaussian filter kernel of each pixel point of each channel image is greater, thereby improving the filtering effect. Finally, the average grayscale index of each super pixel block of each channel image, the larger its value is, the more likely it is a shadow area affected by illumination, and the illumination influence degree of each pixel point obtained in combination with this value is more accurate.

BRIEF DESCRIPTION OF THE DRAWINGS

By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

FIG1 is a flowchart of the steps of a cashmere product color measurement method based on image recognition according to an embodiment of the present invention.

DETAILED DESCRIPTION

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

It should be understood that when the terms "first", "second", etc. are used in the claims, descriptions, and drawings of the present invention, they are only used to distinguish different objects, rather than to describe a specific order. The terms "include" and "comprise" used in the description and claims of the present invention indicate the presence of the described features, wholes, steps, operations, elements, and/or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and/or their collections.

Please refer to FIG1 , which shows a flowchart of a cashmere product color measurement method based on image recognition provided by an embodiment of the present invention. The method comprises the following steps:

S001. Collect the surface image of the cashmere product, and obtain each channel image according to the surface image of the cashmere product.

In an embodiment of the present invention, a camera is used to capture the surface image of a cashmere product. Since the captured image of the cashmere product needs to be used for color measurement, the conventional method cannot be used to grayscale it, and the Gaussian filtering algorithm can only filter two-dimensional images. The captured surface image of the cashmere product is an RGB image, so the Gaussian filtering cannot filter the captured surface image of the cashmere product.

In the embodiment of the present invention, the three channels of the collected cashmere product surface image are separated to obtain each channel image, namely, the R channel image, the G channel image and the B channel image.

S002. Preset the superpixel block size, perform superpixel segmentation on each channel image, and obtain a number of superpixel blocks of each channel image; obtain the average grayscale index of each superpixel block of each channel image; obtain the grayscale sequence of each pixel point of each channel image; obtain the degree of illumination influence of each pixel point of each channel image according to the grayscale sequence of each pixel point of each channel image and the average grayscale index of each superpixel block of each channel image, and then obtain the weight correction coefficient of each pixel point of each channel image.

It should be noted that it is known that the surface of cashmere products is often not smooth enough and is affected by light, which causes grayscale changes in the surface image of cashmere products. In the process of using the Gaussian filter algorithm to filter each channel image to reduce or eliminate the influence of light, the weight assigned to each pixel by the Gaussian filter kernel is only related to the Euclidean distance between the pixels, and the degree to which the pixel is affected by light is not taken into account. Therefore, it is impossible to achieve a good effect of removing the influence of light. Therefore, it is necessary to first obtain the degree of light influence of each pixel in each channel image, and then adaptively obtain the Gaussian filter kernel of each pixel to perform Gaussian filtering on each channel image.

It should be further explained that different areas on the surface of cashmere products are affected differently by light, so it is necessary to segment the channel image and obtain several super-pixel blocks of each channel image for analysis.

In the embodiment of the present invention, the preset super pixel block size m=100. In other embodiments, the implementer may preset the value of the super pixel block size m according to the specific implementation situation, and obtain the size of the nth channel image, which is recorded as M. N, the number of superpixel blocks in the nth channel image at this time , according to the number of super pixel blocks of the n-th channel image, use the super pixel segmentation algorithm to perform super pixel segmentation on the n-th channel image to obtain several super pixel blocks of the n-th channel image; Represents the ceiling function.

It should be noted that it is known that the grayscale values of pixels in the shadow area generated in the channel image due to the influence of light are often low. Therefore, the average grayscale index of each superpixel block in the channel image can be obtained according to the grayscale value of each superpixel block in the channel image. The smaller the grayscale value of any superpixel block in the channel image, the more likely it is to be a shadow area affected by light.

In this embodiment of the present invention, the average grayscale index of the i-th super pixel block of the n-th channel image is obtained:

;

In the formula, Represents the average grayscale index of the i-th superpixel block of the n-th channel image; Represents the average gray value of the pixels in the i-th superpixel block of the n-th channel image; Represents the maximum value of the average grayscale value of pixels in all superpixel blocks of the nth channel image; exp() represents an exponential function with a natural constant as the base; The larger the value of , the more likely the pixel in the i-th superpixel block of the n-th channel image is to be a shadow area affected by lighting.

It should be noted that it is known that the larger the value of the average grayscale index of any superpixel block in the channel image, the more likely the superpixel block is to be a shadow area affected by illumination, that is, each pixel in the superpixel block is more likely to be a shadow area affected by illumination. Furthermore, for any pixel in the channel image, the complexity of the grayscale value distribution of the pixel in its local window can reflect whether the pixel is a pixel in the shadow area affected by illumination. Therefore, the local window of each pixel of each channel image is first obtained, and the degree of illumination influence of each pixel of each channel image is obtained according to the complexity of the grayscale distribution of the pixel in the local window of each pixel of each channel image and the average grayscale index of the superpixel block where each pixel is located.

In the embodiment of the present invention, a local window of each pixel point of the nth channel image is obtained: the local window size is preset to be , with the jth pixel of the nth channel image as the center pixel to construct a pixel with a size of As the local window of the jth pixel of the nth channel image; sort all the pixels in the local window of the jth pixel of the nth channel image in ascending order of grayscale value to obtain the grayscale sequence of the jth pixel of the nth channel image; in the embodiment of the present invention, the local window size is preset In other embodiments, the implementer may preset the local window size according to the specific implementation situation. The value of .

Get the degree of illumination influence of each pixel in the nth channel image:

;

In the formula, Represents the degree of illumination influence of the j-th pixel of the n-th channel image; Represents the average grayscale index of the superpixel block of the n-th channel image where the j-th pixel of the n-th channel image is located; The number of pixels in the local window representing the jth pixel of the nth channel image; Represents the grayscale value of the vth pixel in the grayscale sequence of the jth pixel of the nth channel image; Represents the gray value of the v+1th pixel in the gray sequence of the jth pixel of the nth channel image; The larger the value of , the lower the gray value of the j-th pixel point in the n-th channel image in the superpixel block of the n-th channel image, which means that the gray value in the superpixel block is greatly affected by the shadow caused by illumination, that is, the higher the average gray index of the superpixel block, the higher the degree of illumination influence; Represents the complexity of the grayscale distribution in the local window of the j-th pixel of the n-th channel image. The larger the value, the more complex the grayscale distribution in the local window of the j-th pixel of the n-th channel image, and the greater the degree of illumination influence on the j-th pixel of the n-th channel image.

The illumination influence degree of each pixel point of the n-th channel image is linearly normalized to obtain the weight correction coefficient of each pixel point of the n-th channel image.

S003. Adaptively obtain the Gaussian filter kernel of each pixel of each channel image according to the weight correction coefficient of each pixel of each channel image, perform Gaussian filtering on each channel image, and obtain each channel filtering result image for color measurement.

It should be noted that the weight correction coefficient of each pixel of the channel image reflects the credibility of each pixel during the Gaussian filtering process. The more credible the pixel, the more likely it is to be a pixel with normal exposure. Filtering based on pixels with high credibility can achieve better filtering effects, so that subsequent color measurement results are more accurate. Therefore, it is necessary to correct the weight of each pixel in the Gaussian filter kernel according to the weight correction coefficient of each pixel of the channel image, and adaptively obtain the Gaussian filter kernel of each pixel to perform Gaussian filtering on the channel image.

It should be further explained that in the process of Gaussian filtering, the degree of illumination influence of each pixel is mainly considered. However, in the process of filtering with a fixed window size, when filtering in an area with large illumination shadows, the filtering result will still not be accurate because all the pixels in the window belong to the illumination shadow area and there are no pixels with normal exposure as a reference. Therefore, this problem can be solved by increasing the filtering window and filtering multiple times.

In the embodiment of the present invention, the preset Gaussian filter window size is and filtering times d, obtaining the weight of each pixel in the Gaussian filter kernel of each pixel of the n-th channel image according to the Gaussian function, and multiplying the weight of each pixel in the Gaussian filter kernel of each pixel of the n-th channel image by the weight correction coefficient of each pixel of the n-th channel image, obtaining the updated weight of each pixel in the Gaussian filter kernel of each pixel of the n-th channel image, obtaining the Gaussian filter kernel of each pixel of the n-th channel image, filtering the n-th channel image using the Gaussian filter algorithm according to the Gaussian filter kernel of each pixel of the n-th channel image, stopping when the filtering times of the n-th channel image reaches d, and obtaining the n-th channel filtering result image, in an embodiment of the present invention, the Gaussian filter window size is preset , the number of filtering times d=10. In other embodiments, the implementer can preset the Gaussian filter window size and the number of filtering times according to the specific implementation method; similarly, each channel image is filtered to obtain a filtering result image of each channel.

A multi-channel image mixing algorithm is used to fuse the three channel filtering result images into a color image, and color measurement is performed based on the color image. It should be noted that the multi-channel image mixing algorithm is a prior art and will not be described in detail in the embodiments of the present invention.

The purpose of the present invention is to obtain the degree of illumination influence of each pixel point of each channel image, correct the weight of each pixel point in the Gaussian filter kernel of each pixel point of each channel image, adaptively obtain the Gaussian filter kernel of each pixel point of each channel image to filter each channel image, eliminate the influence of illumination, and make the filtering result of the image more accurate.

The above-mentioned device also includes other components well known to those skilled in the art, such as a communication bus and a communication interface, whose configuration and functions are known in the art and thus will not be described in detail here.

In the present invention, the aforementioned memory may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium may be any suitable magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory, a dynamic random access memory, a static random access memory, an enhanced dynamic random access memory, a high bandwidth memory, a hybrid storage cube, etc., or any other medium that can be used to store the required information and can be accessed by an application, a module, or both. Any such computer storage medium may be part of the device or accessible or connectable to the device.

Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims (6)

1. A cashmere product color measurement method based on image recognition, characterized in that it comprises: Collecting a surface image of a cashmere product; acquiring an image of each channel according to the surface image of the cashmere product; Perform superpixel segmentation on each channel image to obtain multiple superpixel blocks of each channel image; obtain the average grayscale index of each superpixel block of each channel image ; In the formula, Represents the average grayscale index of the i-th superpixel block of the n-th channel image; Represents the average gray value of the pixels in the i-th superpixel block of the n-th channel image; Represents the maximum value of the average grayscale value of pixels in all superpixel blocks of the nth channel image; exp() represents an exponential function with a natural constant as the base; Get the degree of illumination influence of each pixel in each channel image: ,in, Represents the degree of illumination influence of the jth pixel of the nth channel image, Represents the average grayscale index of the superpixel block of the nth channel image where the jth pixel of the nth channel image is located; The number of pixels in the local window representing the jth pixel of the nth channel image; Represents the grayscale value of the vth pixel in the grayscale sequence of the jth pixel of the nth channel image; Represents the gray value of the v+1th pixel in the gray sequence of the jth pixel of the nth channel image; The illumination influence degree of each pixel point of each channel image is linearly normalized to obtain the weight correction coefficient of each pixel point of each channel image; the preset Gaussian filter window size is , according to the Gaussian function, the weight of each pixel in the Gaussian filter kernel of each pixel of each channel image is obtained, and the weight of each pixel in the Gaussian filter kernel of each pixel of each channel image is multiplied by the weight correction coefficient of each pixel of each channel image, and the updated weight of each pixel in the Gaussian filter kernel of each pixel of each channel image is obtained to obtain the Gaussian filter kernel of each pixel of each channel image; based on the Gaussian filter kernel of each pixel of each channel image, each channel image is Gaussian filtered to obtain the filtering result image of each channel. 2. The cashmere product color measurement method based on image recognition according to claim 1, characterized in that the super-pixel segmentation of each channel image to obtain a plurality of super-pixel blocks of each channel image comprises: Preset the superpixel block size m and obtain the size of the nth channel image, denoted as M N, then the number of superpixel blocks in the nth channel image , according to the number of super-pixel blocks of the n-th channel image, use the super-pixel segmentation algorithm to perform super-pixel segmentation on the n-th channel image to obtain several super-pixel blocks of the n-th channel image; Represents the ceiling function. 3. The cashmere product color measurement method based on image recognition according to claim 1, characterized in that the acquisition of the local window comprises: The default local window size is , with each pixel of each channel image as the center pixel to construct a pixel with a size of The local window of is used as the local window of each pixel point of each channel image. 4. The cashmere product color measurement method based on image recognition according to claim 1, characterized in that the acquisition of the grayscale sequence comprises: All the pixels in the local window of each pixel of each channel image are sorted in ascending order of grayscale value to obtain the grayscale sequence of each pixel of each channel image. 5. The cashmere product color measurement method based on image recognition according to claim 1, characterized in that the Gaussian filter kernel based on each pixel point of each channel image performs Gaussian filtering on each channel image to obtain each channel filtering result image, including: The number of filtering times d is preset, and the images of each channel are filtered using the Gaussian filtering algorithm according to the Gaussian filtering kernel of each pixel point of each channel image until the number of filtering times of each channel image reaches d, thereby obtaining the filtering result images of each channel. 6. A cashmere product color measurement device based on image recognition, characterized in that it includes: a processor and a memory, the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the cashmere product color measurement method based on image recognition according to any one of claims 1 to 5 is implemented.