A Blurred Image Retrieval System Research And Implementation

Image retrieval is a key issue in the field of computer vision, and has a wide application in the Internet business model. In the present, the inquire images and database images for an image retrieval system are all clear images, and then the inquire images can be recognized with an image matching process. However, in real applications images might be blurred due to noise in the image acquisition, storage and transmission process. Since the reason and parameters of image blurring are usually unknown, it is difficult to restore or reconstruction a blurred image. As a result, the blurred image cannot be recognized in the existing image retrieval system. In this paper, an image retrieval system for blurred images is investigated and developed.The basic idea of blurred image retrieval is to expand the image database by creating more database images. For image feature selection, to use the SIFT and color histogram features is a common practice. However, it is difficult to accurately describe the characteristics of an image if only one image feature is used. Therefore, this paper combines the SIFT and color histogram features, and proposes an effective recognition algorithm for blurred images. In order to identify the blurred image, we investigated a strategy to expand the image database, which simulate the process how blurred images could be created. The blurred image space is constructed via the following methods: the effect of image blurring via image noise is simulated with the Gaussian filter; the effect of image blurring due to illumination is simulated with an adjustment of image histogram; the effect of image blurring due to motion is simulated with a motion filter. Since the construction of the blurred image space can be performed offline, the online computational costs only increase slightly. Experimental results are presented to show the efficiency of the new algorithm.