Image Retrieval Algorithm Based On Multi-feature And Implementation

Content-based image retrival is an important research focus in the field of computervision, It is the development and application of scientific and technological progress. Thetechnique use the image of its own color, texture, shape, and spatial position relationshipbetween the low-level visual features to describe the content of the image characteristics,and then to measure the similarity between images according to the distance between theimage feature vector, thereby completing the image retrieval. It can be seen, accuratelyand efficiently extract image low-level visual features are extremely important in CBIR.Inrecent years,image retrival technology based on a single feature already have somedevelopment.but single feature can not represent the image content, therefore, acombination of more features to improve the retrieval accuracy become available method.But how to effectively combine a variety of different features to complement each other toa complete representation of the image content, and retrieve images consistent with humanvisual perception is a problem worthy of further study.We studied the algorithm for how to effectively extract the image’s features based onthe visually important image edge contour. Improved color and shape feature extractionalgorithm according to the multi-scale characteristics of the wavelet transform, andproposed image retrieval new algorithm based on color edge multi-feature. Edge qualitywill affect the retrieval efficiency, the dyadic wavelet, range between continuous waveletand discrete wavelet, is a wavelet function with a strong directionality, translationalinvariance, can effectively avoid the visual deformation caused by the non-lineartransformation, has some anti-noise capability which makes it wide range of applicationsin image singularity detection and feature extraction. So in this paper we choose theexcellent performance good binary wavelet filter, and then extracted the image color edgethrough the dyadic wavelet transform modulus maxima method. For the color featureextraction, choose the center of gravity of color edge point collection as the center of thecircle，and divided the image space into a series of concentric rings according to thedistribution of the edge points. Eventually we can get the annular color histogram throughcalculating the color histogram of the area of each ring. Annular color histogram has astrong robustness to the image illumination change and geometry change and overcomethe shortcomings of lack of spatial information of the traditional color histogram. For theshape feature extraction, the same way image space were divided into a series ofconcentric rings,and calculate wavelet modulus maxima of edge points. Then constructwavelet modulus maxima annular histogram. Statistics to the direction of the gradient of the edge points to get the edge gradient direction histogram, and the normalizationprocessing, can be obtained an image shape characteristics.Finally, comprehensive utilize the above three kinds of histogram which fully reflectthe edge contour information to describe image content features, and conducted imageretrieval experiment in the public standard image database. The experimental results showthat the extracted feature vector contains a variety of naturally underlying information ofthe image, thereby fully complementary showing the image content. and compared withthe similar algorithm,we know that the algorithm not only has better precision and recallrate, but also to image illumination variations and geometric changes (scale, translation,rotation, etc.) has a strong robustness.