CN103578123B - A kind of image-region merges method - Google Patents

Abstract

The present invention relates to a method for merging image areas, characterized in that: the color histogram distance between image area A and image area B is obtained by the following method, in, In the formula, p(k) is the color histogram of image area A, q(k) is the color histogram of image area B, where k=0,1,...,N _color -1, N _color represents the color histogram The number of color quantization regions used, [A ₀ , A ₁ ] is the support region of p(k), [B ₀ , B ₁ ] is the support region of q(k), and D(p,q) represents two images The distance between the region color histograms; according to the size of the color histogram distance D(p,q), it is judged whether to merge the image region A and the image region B.

Description

A method of image region merging

technical field

The invention relates to a method for merging image regions.

Background technique

Image segmentation plays an important role in pattern recognition and image processing. The results of image segmentation directly determine the performance of image compression techniques for object tracking, image understanding, and objects of opportunity. At present, many image processing methods have been proposed, such as: threshold segmentation method, feature space clustering technology, boundary-based segmentation method, region-based segmentation method, graph segmentation method and so on. In recent years, graph segmentation methods have attracted more and more attention from researchers. The graph segmentation method mainly studies the node clustering problem, and the mainstream methods include Normalized Cut (N-Cut) and Ratio Cut (RatioCut, R-Cut). However, the graph segmentation method has two disadvantages: the required storage space is large, and the calculation speed is slow. To overcome these two shortcomings, researchers propose a multi-level image segmentation method that combines region-based image segmentation methods with graph segmentation methods. For example: Yang et al. proposed an image segmentation method based on watershed square and graph theory; Tao et al. proposed an image segmentation method based on mean shift (MeanShift, MS) and N-Cut.

Although these region-based image segmentation methods and multi-level image segmentation methods of graph segmentation have great advantages in overcoming storage space and computing speed, they usually use simple features (such as color mean) to characterize image regions. Because in multi-level image segmentation, the image area obtained after the primary segmentation of the image is usually relatively large, the features of the image area cannot be fully represented by simple feature vectors. The research shows that: the color histogram can represent the global and local characteristics of the image color distribution. Therefore, the color histogram of an image can represent more features. However, Euclidean distance, histogram intersection distance, or Bhattacharyya distance, which are suitable for pattern matching and classification, are not suitable for image region merging.

Contents of the invention

The purpose of the present invention is to provide a method for merging image regions, which can effectively realize the merging of image regions.

Realize the technical scheme of the present invention:

A method for merging image regions, characterized in that:

The color histogram distance of image area A and image area B is obtained by the following method,

in,

In the formula, p(k) is the color histogram of image area A, q(k) is the color histogram of image area B, where k=0,1,...,N _color -1, N _color represents the color histogram The number of color quantization regions used, [A ₀ , A ₁ ] is the support region of p(k), [B ₀ , B ₁ ] is the support region of q(k), and D(p,q) represents two images distance between region color histograms;

According to the size of the color histogram distance D(p,q), whether to merge the image area A and the image area B is judged.

When the color histogram distance D(p,q)<a certain threshold, the image area A and the image area B may be merged.

The beneficial effect that the present invention has:

The color histogram distance calculation method proposed by the present invention fully considers the characteristics of image regions in multi-level image segmentation, has a high accuracy of image region merging, and is simple and easy to implement.

Description of drawings

Fig. 1 is embodiment image;

Fig. 2 is a color histogram of image regions A, B, and C in Fig. 1 .

detailed description

Let A and B denote two image regions, p(k) is the color histogram of region A, q(k) is the color histogram of region B, where k=0,1,...,N _color -1, N _color Indicates the number of color quantization areas used in the color histogram, [A ₀ ,A ₁ ] is the support area of p(k), that is: p(k)≠0 if k∈[A ₀ ,A ₁ ], [ B ₀ , B ₁ ] is the support area of q(k), namely: q(k)≠0 if k∈[B ₀ , B ₁ ]. Define the histogram distance of regions A and B as:

in,

In formula (1), D(p,q) represents the distance between the histograms p and q, and ∩ represents the intersection of sets.

Formula (1) shows that: 1) If region A contains all colors that appear in region B, then the histogram distance between A and B is 0; 2) similarly, if region B contains all colors that appear in region A color, the histogram distance between A and B is also 0; 3) If the region A does not contain any color in region B, the histogram distance between A and B is also 1; 4) If it is not the above 1 ), 2), and 3), the histogram distance between A and B is The values of K ₀ and K ₁ are related to the intersection of the two sets, and are determined by formula (2).

The beneficial effects of the present invention will be further described below in conjunction with specific examples.

Fig. 1 shows an image and its three regions A, B and C, and Fig. 2 shows the color histograms of regions A, B and C, where N _color =512. The histogram distance between regions A and B calculated by formula (1) is 0.4912, the histogram distance between regions A and C is 0.0795, and the histogram distance between regions B and C is 0.9635. From this, we can judge that the colors of regions A and C are very close and can be merged into one region. This is consistent with the clustering results of the human eye.

On the other hand, if the traditional Bhattacharya distance is used, the histogram distance of the regions A, B is 0.9995, the histogram distance of the regions A, C is 0.6472, and the histogram distance of the regions B, C is 0.9993. The histogram distance of regions A and C is 0.6472, indicating that the colors of these two regions are not similar and should not be merged into one region, which obviously does not match the classification results of the human eye.

Claims (1)

1. A method for merging image regions, characterized in that: The color histogram distance of image area A and image area B is obtained by the following method, in, In the formula, p(k) is the color histogram of image area A, q(k) is the color histogram of image area B, where k=0,1,...,N _color -1, N _color represents the color histogram The number of color quantization regions used, [A ₀ , A ₁ ] is the support region of p(k), [B ₀ , B ₁ ] is the support region of q(k), and D(p,q) represents two images Distance between region color histograms: 1) If region A contains all colors that appear in region B, then the histogram distance between A and B is 0; 2) If region B contains all colors that appear in region A color, the histogram distance between A and B is also 0; 3) If region A does not contain any color in region B, then the histogram distance between A and B is 1; 4) If not the above 1) , 2), and 3), the histogram distance between A and B is According to the size of the color histogram distance D(p,q), whether to merge the image area A and the image area B is judged.