CN110647910A - Image similarity calculation method based on color quantization - Google Patents

Abstract

The invention discloses an image similarity calculation method based on color quantization, which comprises the following steps: step 1): converting the value of the HSV color space: step 2): carrying out quantitative dimensionality reduction on the converted HSV color space; step 3): given image P to be matched_dAnd P_c(ii) a Step 4): image P_dAnd P_cEqually dividing into n parts from top to bottom, step 6): deriving an image P from a color quantized component map_dkIs marked as W_dk，k＝1,2,3,...,n，W_ckK is 1,2,3,. cndot, n; step 8): given a threshold T, calculate the color W_dkWhether or not it is equal to the color W_ckK is 1,2, 3. The invention utilizes simple and effective methods in the fields of computer vision and image over-processing to carry out image similarity calculation, utilizes strategies of color space quantization and dimension reduction, not only improves the accuracy of image matching during similarity calculation, but also greatly savesThe time cost is reduced, the real-time requirement on the speed is met, and the method can be applied to various fields needing to calculate the similarity in real time.

Description

Image similarity calculation method based on color quantization

Technical Field

The invention relates to the fields of computer vision, image processing and the like, in particular to an image similarity calculation method based on space color quantization.

Background

With the progress of social science and technology, the computer vision technology and the image processing field are rapidly and vigorously developed, and the technology of image similarity matching as an important component of each field of image processing has long become an indispensable technology for the computer vision and the image processing direction, and is applied to each existing or emerging field.

Photo album clustering, face recognition matching, pedestrian re-recognition, vehicle re-recognition and the like, and the image similarity calculation cannot be carried out in the emerging computer vision field which is more hot in recent years. Especially in recent years, with the rise of deep learning, the similarity calculation of images is more hot. The deep learning method has high precision and high accuracy which are not possessed by the traditional image method, and is not only reflected in the aspect of image similarity calculation, but also reflected in the aspect of the computer vision field. For example, classification and identification of objects, detection and tracking of objects, and the like, so that the similarity of images is mostly calculated by adopting a deep learning method in the current emerging field. However, deep learning also has a large disadvantage, and a large amount of data and a large amount of time are required to obtain a final matching result, which is difficult to apply in some real-time image matching and similarity calculation fields.

Therefore, it is very meaningful to be able to complete image similarity matching very accurately with a very small time overhead by using a simple computer vision and image processing method.

Wuhan university, 2014 Zhanghua, a method for calculating image similarity through object position information in an image, and meanwhile, information of a color histogram is introduced for more accurate calculation. The similarity of the images is calculated by performing feature point matching and purification by using a violence algorithm and a random sampling consistency algorithm, wherein the similarity mainly aims at pedestrian images, and feature point extraction and description are performed on multi-scale images by using an improved FAST algorithm and a BRIEF algorithm. The south China university of science and engineering, 2015 Qiyuhui and the like, provides a characteristic distance ordering mode combining measurement learning and sparse representation methods for image matching and similarity calculation, converts extracted image characteristics according to the semi-positive nature of a measurement matrix in the mahalanobis distance to obtain new characteristics, integrates the new characteristics into a sparse model to evaluate the similarity between a test sample and a dictionary, orders individuals with non-zero sparse coefficients in a data set according to standard reconstruction errors by using an iterative sparse coefficient reweighing mode, and judges the similarity between images according to the ordered final result. The university of east China, 2017 Liuna and the like, provides an improved Simese structure based deep convolutional neural network for image matching and similarity calculation. During training, classification and similarity measurement are combined, so that the distance between classes is increased, the distance in the classes is reduced, effective features in the images are extracted, and then the similarity between the two images is calculated by further using a metric learning algorithm. Nanjing post and telecommunications university, 2017 Tangsong, proposes an image similarity calculation method based on salient features, utilizes superpixels to construct a feature space, utilizes a method based on cellular automata to calculate the intrinsic salient features of images, and utilizes a learning ordering method to calculate the similarity between the images.

Although the above documents and methods all refer to image similarity calculation and matching by computer vision and the like, the following disadvantages still exist:

(1) the time cost is too large, a large amount of time is needed for calculating the similarity of the two images, and the method is difficult to play a role in some occasions needing real-time calculation;

(2) the accuracy of similarity matching is not high, the existing similarity calculation method is basically applied to various matching and re-identification fields, but the accuracy of the similarity calculation method caused by the difference between images is not high;

therefore, how to perform image similarity calculation and image matching by a simple and effective method so as to quickly and accurately apply image similarity matching in an application scene in a real-time environment is a problem which needs to be solved at present.

Disclosure of Invention

In order to overcome the defects of the algorithm and the method and improve the efficiency and the accuracy of image similarity calculation, the invention provides an image similarity calculation method based on color quantization.

The technical scheme of the invention is as follows:

an image similarity calculation method based on color quantization is characterized by comprising the following steps:

step 1): the values for the HSV color space are transformed according to the following equation:

wherein H₀，S₀，V₀Value, H, representing the image in the original HSV space₁，S₁，V₁Representing the converted value;

step 2): carrying out quantitative dimensionality reduction on the converted HSV color space;

step 3): given image P to be matched_dAnd P_c；

Step 4): image P_dIs divided into n parts from top to bottom, and is marked as a set C_d＝{P_dk|k＝1,2,3,...,n}；

Step 5): also image P_cIs divided into n parts from top to bottom, and is marked as a set C_c＝{P_ck|k＝1,2,3,...,n}；

Step 6): traverse set C_dMiddle image block P_dkCalculating HSV quantitative components of all the pixel points and averaging to obtain an image P_dkOf HSV

Deriving an image P from a color quantized component map_dkIs marked as W_dk，k＝1,2,3,...,n；

Step 7): traverse set C_cMiddle image block P_ckCalculating HSV quantitative components of all the pixel points and averaging to obtain an image P_ckOf HSV

Deriving an image P from a color quantized component map_dkIs marked as W_ck，k＝1,2,3,...,n；

Step 8): given a threshold T, calculate the color W_dkWhether or not it is equal to the color W_ckN, where N is the number of equal colors, and if N is equal to or greater than T, the image P is determined to be a target image P_dAnd P_cSimilarly; otherwise, it is determined to be dissimilar.

The invention has the advantages that: the image similarity calculation is carried out by a simple and effective method in the fields of computer vision and image over-processing, and the strategies of color space quantization and dimension reduction are utilized, so that the accuracy of image matching is improved when the similarity calculation is carried out, the time overhead is greatly saved, the real-time requirement is met in the speed, and the method can be applied to various fields needing to calculate the similarity in real time.

Drawings

FIG. 1 is a diagram of HSV space color quantization;

FIG. 2 is an input image to be matched;

FIG. 3 is an image to be searched;

FIG. 4 is a segmentation diagram of an image to be matched;

fig. 5 is a segmentation diagram of an image to be searched.

Detailed Description

The following describes a specific embodiment of the image similarity calculation method based on color quantization according to the present invention based on an example.

Step 1): the values for the HSV color space are transformed according to the following equation:

wherein H₀，S₀，V₀Value, H, representing the image in the original HSV space₁，S₁，V₁Representing the converted value;

step 2): carrying out quantitative dimensionality reduction on the converted HSV color space; in the present embodiment, the colors are divided into ten colors of black, gray, white, red, orange, yellow, green, cyan, blue and purple, which are corresponded to fig. 1;

step 3): given image P to be matched_dAnd P_c(ii) a In the present embodiment, the image P_dAnd P_cAs shown in fig. 2 and 3;

step 4): image P_dIs divided into n parts from top to bottom, and is marked as a set C_d＝{P_dk1, | k ═ 1,2,3,.., n }; in this embodiment, the value of n is 6, and the specific division result is shown in fig. 4;

step 5): also image P_cIs divided into n parts from top to bottom, and is marked as a set C_c＝{P_ck1, | k ═ 1,2,3,.., n }; in this embodiment, the value of n is 6, and the specific division result is shown in fig. 5;

step 6): traverse set C_dMiddle image block P_dkCalculating HSV quantitative components of all the pixel points and averaging to obtain an image P_dkOf HSV

Deriving an image P from a color quantized component map_dkIs marked as W_dkK is 1,2,3,. cndot, n; in this embodiment, n is 6;

step 7): traverse set C_cMiddle image block P_ckFor each pixel of (1), calculating all pixelsHSV quantization component and averaging to obtain an image P_ckOf HSV

Deriving an image P from a color quantized component map_dkIs marked as W_ckK is 1,2,3,. cndot, n; in this embodiment, n is 6;

step 8): given a threshold T, calculate the color W_dkWhether or not it is equal to the color W_ckN, where N is equal to N, and if N is equal to or greater than T, the image P is determined to be a target image P_dAnd P_cSimilarly; otherwise, judging that the two are not similar; in this embodiment, n is 6, T is 3, and if the condition is satisfied, the two images are considered to be similar, otherwise, the two images are considered to be dissimilar.

The embodiments described in this specification are merely illustrative of implementations of the inventive concept and the scope of the present invention should not be considered limited to the specific forms set forth in the embodiments but rather by the equivalents thereof as may occur to those skilled in the art upon consideration of the present inventive concept.

Claims (1)

1. An image similarity calculation method based on color quantization is characterized by comprising the following steps:

step 1): the values for the HSV color space are transformed according to the following equation:

wherein H₀，S₀，V₀Value, H, representing the image in the original HSV space₁，S₁，V₁Representing the converted value;

step 2): carrying out quantitative dimensionality reduction on the converted HSV color space;

step 3): given image P to be matched_dAnd P_c；

Step 4): image P_dIs divided into n parts from top to bottom, and is marked as a set C_d＝{P_dk|k＝1,2,3,...,n}；

Step 5): also image P_cIs divided into n parts from top to bottom, and is marked as a set C_c＝{P_ck|k＝1,2,3,...,n}；

Step 6): traverse set C_dMiddle image block P_dkCalculating HSV quantitative components of all the pixel points and averaging to obtain an image P_dkOf HSV

Deriving an image P from a color quantized component map_dkIs marked as W_dk，k＝1,2,3,...,n；

Step 7): traverse set C_cMiddle image block P_ckCalculating HSV quantitative components of all the pixel points and averaging to obtain an image P_ckOf HSV

Deriving an image P from a color quantized component map_dkIs marked as W_ck，k＝1,2,3,...,n；

Step 8): given a threshold T, calculate the color W_dkWhether or not it is equal to the color W_ckN, where N is the number of equal colors, and if N is equal to or greater than T, the image P is determined to be a target image P_dAnd P_cSimilarly; otherwise, it is determined to be dissimilar.

Images