A Histogram Similarity Measurement Method

Abstract

The invention belongs to the technical field of image processing, and particularly relates to a histogram similarity measuring method based on hierarchical matching. The histogram similarity measuring method comprises the steps of histogram normalization, histogram classification, classification similarity judgment and histogram similarity measurement. The histogram similarity measuring method well overcomes the defects that in existing histogram matching, shifting is sensitive, robustness is weak and accuracy is not high, is good in robustness and high in accuracy, and establishes solid foundation for further operation based on histogram matching.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a histogram similarity measurement method based on hierarchical matching. Background technique [0002] Histogram is a very important analysis tool in image processing, which can reflect the distribution of global information such as image grayscale, texture, and color. It is widely used in image matching, target tracking, audio processing and other fields because of its simple calculation and invariance to scale, translation, and rotation of images. The matching accuracy between histograms will directly affect the correctness of subsequent operations. The existing histogram similarity measurement methods mainly include distance measurement methods, such as Manhattan distance, Euclidean distance, Hausdorff distance, central moment method, x 2 Statistical distance, etc.; correlation calculation methods, such as cosine correlation, Pearson product-moment correlation, hist...

AI Technical Summary

Problems solved by technology

Whether it is to find the distance between points in the feature space or the correlation between feature vectors, only the values ​​​​of the corresponding columns of the two histograms are calculated, and the distribution of the histograms is not considered. Therefore, when the histogram is affected by noise or illumination, etc. When the graph is shifted and deformed, these metrics will not be able to accurately calculate its similarity

James Hafner et al. proposed the quadratic form distance in the article "Efficient color histogram indexing for quadraic form distance functions". By using a similarity matrix to add cross-column information, the matching between different columns of two histograms is realized. However, due to the The method will overestimate the similarity of feature distribution and lead to low matching accuracy

Madirakshi Das et al. proposed the peak matching method in the article "Searching for multi-colored objects in a diverse image database", but this method usually has a better matching effect on specific images in specific applications, and is not universal enough

Yossi Rubner et al. proposed the EMD (earth mover's distance) method in the article "The earth mover's distance as a metric for image retrieval". It is not sensitive to the small shift of the histogram, but it is difficult to apply to the histogram with too few and too few bars. In the case of too many columns, the "ground distance" cannot be obtained when there are too few columns, and the calculation is slow when there are too many columns, and the optimal allocation problem must be considered in the matching process, which is highly complex

Images