Image Retrieval Based On The Histogram Of Oriented Gradient Of Interest Points

With the rapid development of Internet industry especially themultimedia technology and the combination of traditional industries andInternet technology, a large number of image and video resources arecoming out. In the mean time, these resources bring much convenience topeople’s daily work and life. How to better restore、retrieve and makegood use of the expanding image database is becoming one of thetechnologies that many scholars focus on recently. This paper mainly dosome research on image retrieval. In order to improve the speed andprecision of image retrieval, this paper makes the image detail more clearthrough image gray-scale enhancement before extracting interest points.Then based on interest points and the neighborhood, divide the image intofour parts, calculate the histograms of oriented gradient for every part asthe feature vectors and do similarity measure at the end. The main worksare as follows: (1) Image color spaces are studied especially the HSV color spaceand we mainly focus on non-uniform equalization of HSV color spacewhich has been widely used in image retrieval. Finally, experiments areperformed based on these features.(2) Image visual features such as color, texture, shape and spacefeatures are studied. This paper mainly introduces histogram of imagecolors and Gabor wavelets, then describe image features with these twofeatures. Finally, experiments are performed.(3) Before extracting interest points, the color images need to betransformed to gray-scale ones. In this process, some detail features maybe lost around the areas which are sad colored or the areas which hassingle color. To make sure that we can extract enough points to get themore precise results, image enhancement methods of spatial domain andfrequency domain are studied. And this paper adopts histogramequalization. The number of extracted interest points are comparedbetween the enhanced images and the non-enhanced images with thesame extracting method.(4) Harris corner detector, scale invariant feature transform andspeed up robust features are studied respectively. The advantages anddisadvantages of these three algorithms are compared and theimprovements of these three algorithms are briefly introduced. Becauseof the advantages of speed up robust features, it is adopted in this paper and some experiments about extracting interest points are performedbased on this method.(5) Image retrieval method based on the histogram of orientedgradient of interest points is proposed in this paper. Based on the idea ofweight, the images are divided into four parts and the possibility of thenumber of interest points in each part is regarded as the weight value ofeach part. The histogram of oriented gradient of each part is multiplied bythe weight value of the according part to get the final histogram. Thus,the images are represented with a feature vector of64dimensions. Thisalgorithm made good use of the space features of an image and theexperiment results show its effectiveness.