Based On The Webcam Photo Retrieval Technology

With the fast popularizing of digital cameras and rapid increasing of storage capacity, it is not uncommon to store thousands of images on a computer. Content based image retrieval (CBIR) is therefore becoming a popular research field in all kinds of academic or commercial institutions. In most of CBIR systems, query image is usually obtained from the same image database that is being retrieved. This is least useful in some practical cases if users are looking for some specific images in an image database. For example, user may want to reprint some photos from a large database. He/she has the previously printed photo on hand and wants to find the same photo in database and reprint more copies. In this scenario, the best query image is on user's hand and the easiest way to input the image is using an online video camera.This paper is to introduce a new technique for the real-world photo searching. With an average Webcam, the technique is to have a snapshot at a printed photo, then automatically find out the most similar photos, especially the original photo of the print, in the digital photo storage. It includes three technical parts.Part 1 is Content Based Image Retrieval. We extract color, edge, and texture feature from images. With these features we can define the similarity between images and find the most matching ones in the image database. In the retrieval result, users can specify those images that are not similar with what they want. Based on this information, the system can automatically adjust the importance of each feature and give an improved retrieval result.Part 2 is Snapshot Processing. The main role of this part is to extract the photo from the snapshot and make it the example for retrieval. It should be able to deal with blurred, tilted, and incompletely viewed images in real world.Part 3 is the Color Constancy problem. In changed lighting condition, the color of the snapshot will also change, which influenced the effect of the retrieval result. Therefore, we must find the constant colors which are robot to the change of the light and all possible external factors. By analysing and testing a quantity of appropriate algorithms, we finally determined an optimal solution.Our experimental results showed that we can achieve satisfactory results, which means our algorithms used in this technique are successful.