Similar Image Retrieval Using Color Histogram In HSV Space And SIFT Descriptor

The development of multimedia technology, produces a large number of pictures in the Internet, and makes the image retrieval more and more important. The complexity and subjectivity make the traditional text-based image retrieval technique can’t meet the application requirements, thus giving birth to content-based image retrieval technique. In this article, the key techniques of content-based image retrieval are introduced. It is including the feature extraction, similarity matching and the performance measurement of the CBIR.The underlying visual features of the CBIR belong to the global features, and they are usually only a vague description of the similarity between images. SIFT algorithm which is proposed by David Lowe is one of local feature descriptors and is invariant to rotation, scale, illumination changes, viewpoints changes, affine invariant, occlusion and noise. So the SIFT algorithm is stable for similar image retrieval which the images are photographed in the same scene, different viewpoints and different illumination.Base on the SIFT algorithm, the problem which the high dimensionality of the SIFT feature descriptor and the big of the image database brought difficulties for image matching is proposed. To solve this problem, this paper studied today’s popular image retrieval with visual words based SIFT features, and used the idea of TF-IDF and inverted list in the test retrieval to improve the accuracy and efficiency. The similar images in this paper were color images, and SIFT feature only used the gray information in the image. This paper proposed a method that combined color histogram in HSV space and visual words with TF-IDF based SIFT features. Experimental results demonstrate that the performance of this scheme is efficient.