A Fast Image Retrieval Method Based on Color Feature and Texture Feature

Abstract

The invention relates to a quick image retrieval method based on color features and texture characteristics. The method comprises steps of removing a background in a to-be-retrieved image, compressing RGB color space of each image in a database, extracting a histogram of each picture in the database, compressing RGB color space of the to-be-retrieved image, extracting the histogram of the to-be-retrieved image, extracting histogram features of the to-be-retrieved image, conducting rarefaction to the extracted histogram features, calculating the distance of the corresponding histogram features after rarefaction, sorting all images in the database based on histogram feature distance for the first time, selecting the former part of images in the first sorting results to extract texture characteristics, calculating a minimum distance between texture characteristics, and sorting for the second time based on the texture characteristics distance and acquiring the final retrieval results. Large number of image coupling retrieval can be finished in short time by the use of the quick image retrieval method and high accuracy can be achieved.

Description

technical field [0001] The invention relates to a fast image retrieval method based on color features and texture features. Background technique [0002] Image retrieval technology has important application value in search engines, big data mining, and image understanding. Researchers have proposed a series of image retrieval methods from many perspectives such as image signal processing and pattern classification. These methods are based on the understanding of image content to judge the similarity between two images. However, the operation speed of many current retrieval methods is slow, which cannot meet the needs of many practical applications, which limits the development of image retrieval technology. [0003] The understanding of image content can be carried out through the histogram statistics of color or grayscale. This type of method has the advantage of small amount of calculation, but the degree of understanding of the content is limited, only staying at the ove...

AI Technical Summary

Problems solved by technology

[0005] Chinese patent CN104899326A proposes an image retrieval method based on binary multi-index hash technology. This method mainly uses principal component analysis to distinguish the difference and similarity between images. Although the hash index has high computing efficiency, but This method needs to rely on a fixed image database for training, and its general performance is poor. When the image is disturbed by factors such as light, the principal component method cannot correctly judge the similarity between images.

Chinese patent CN104899280A proposes a fuzzy correlation asynchronous image retrieval method based on color histogram and NSCT. This method directly uses the characteristics of the color histogram to distinguish image content. The fuzzy function used depends on empirical data. The retrieval performance The performance is unstable. This method also uses texture features for image retrieval, and its computational efficiency is not high.

Chinese patent CN104850644A proposes a Lucene-based hyperspectral image retrieval method and retrieval engine, which ignores the difference in image content and cannot be used for image content retrieval

Chinese patent CN104834748A proposes an image retrieval method based on deep semantic sorting hash coding. This method uses convolutional neural network to dig deep into image semantics and has high image understanding ability, but its retrieval algorithm wastes Computing resources, in the retrieval of massive images, over-in-depth analysis of images with obvious differences, this method has room for further acceleration

Chinese patent CN104834732A proposes a texture image retrieval method, which uses traditional texture features for matching, and its algorithm is greatly affected by image rotation and geometric deformation, and has a high amount of computation

Images