Image Retrieval Technology Research Based On Color Vector Angle Difference Feature

With the development of the information society, the appearance and widespread use of large capacity storage and digital devices, as well as the rapid popularization of multimedia and network technology, image data presents the growth trend of geometric progression. It is a worthy project for further exploring when users search for images that they need in huge data. Generally, keyword-based search engine lacks of objectivity, and too heavy workload of manual tagging. Recently in the field of multimedia, it has become a hot research issues that how to retrieve images from massive image data quickly and effectively. Therefore, it becomes an urgent need to establish a new retrieval system of semi-automatic or automatic based on multimedia content. Currently, the wide-ranging research of CBIR (Content-Based Image Retrieval) technology aims at this challenge. With the continuous digging of the potential and rapid development of the technology, it processes a high theoretical and practical value, which will provide strong support for the management and access of large-scale image database, as well as widely used in many fields.It found that color processes a lot inherent traits which they have little dependence of the image's size, direction and perspective, as well as rotation invariant, by reading a lot of relevant literature, analyzing and researching different visual feature extraction algorithms. However, most pioneers built feature vectors by extracting frequency statistics of the colors, which lose spatial distribution and location information of the image content. Although its extraction methods are simple, but the effectiveness is unsatisfied.This paper proposed an image retrieval algorithm which adopts color vector angle difference histogram based on the analysis of color feature. The algorithm traverses pixels to acquire color vector angle, and establishes the maximum color vector angle template to get the largest changes pixels, and then carry quantization for the vector angle and color space to get color histogram of and vector angle histogram, combining two clues to establish feature vector index by extracting vector angle difference histogram. Ultimately, it got the optimal dimension reduction parameters of color and angle through a large simulation contrast experiment. Analyzing the measure perspective of different match criteria, the method adopted the optimized match criteria to reflect the difference of feature vectors. The proposed algorithm is not only retain the spatial distribution of colors, but also achieve the edge orientation of the object. Meanwhile, the method doesn't require any color space conversion. Simulation results show that the algorithm has a good retrieval effectiveness.