Research Of Various Feature-Based Image Retrieval In News Photo-Database

Searching a group of similar photos from the database, or checking photos repeated before putting into the database are commonly used in the news photo-database system. The Judgment of the photos similarity heavily depends on human's visual recognitions on photo's feature such as color and structure. However, these visual recognitions are hard to be literarily described. Moreover, literature description of photos is widely different individually which likely causes big disagreement by different individuals' description. Therefore, a image searching system based on the image's various features is developed to eliminate the problems of low efficiency and unsatisfied result caused by keywords based approaches. It brings the function of searching objectively with related photos. Today, the mature CBIR technology mainly bases on underlying feature such as color, texture, and many effective feature descriptors have been gotten, such as color histogram, contrast, coarseness, directionality. The color histogram of HIS model and CLD, EHD in MPEG-7 standard are applied to four areas with different weight to describe the content of the images. According to user's relevant valuation to searching result, we suggest using relevance feedback to get keywords of news object objectively and make the searching result close to user's need. The system shows very good result based on the test from a photo database with 1000 news photos.