After PCA pre-processing, rough set theory was introduced in image's feature attributes reduction, and its application in characterized parameters' attribute optimization was explored. The combination of these two methods was effective in reducing the unnecessary attributes. The novel algorithm could also decrease the complexity of CBIR's inner redundancy. The experimental result of attribute reduction using UCI dataset also indicated that there was in-built redundancy of the original features and the complexity of the follow-up processing had cut down through employing the methods mentioned in this paper.