Research And Implementation Of Image Retrieval System Based On SIFT And Rotation Invariant LBP Combination

Due to its own restrictions, traditional text-based image retrieval can not meet the needs of the users in image information retrieval. A new method of image retrieval-content-based image retrieval emerged. And it has become an important means of efficient management and retrieval of large-scale multimedia database. Therefore, a content-based image retrieval system is designed after deep analysis.Actually, images always get some changes in scale, viewpoint and brightness. However, the traditional image features-color, texture, shape and spatial relationships, that content-based image retrieval extracted, always perform poorly. So, a scale, rotation, translation and even brightness invariant feature extraction algorithm is badly needed. This paper deeply studies the principles and methods of the Scale Invariant Feature Transform (SIFT) algorithm and Local Binary Pattern (LBP) algorithm. For the poor performance of SIFT when image illumination changes, rotation invariant LBP is focused and detailedly study.Then, a new image feature extraction algorithm-SIFT and rotation invariant LBP combination is proposed. It uses the same key-point detection method as SIFT. After getting the key-points of the image features, the SIFT-LBP descriptor is made up of statistics of gradient information in16x16region and rotation invariant LBP value in9x9region around each key-point.In order to further improve the retrieval accuracy and retrieval speed, a feature selection method based on genetic algorithm is selected to remove the redundant feature information. And an index structure based on Bag of features and VP-tree is proposed. With Bag of features model, the high-dimensional feature data of the image can be mapped to the visual vocabulary. And then, the image can be represented by a feature vector based on the number of visual word frequency. Then the index tree of VP-tree is constructed the feature vectors of all images. Under the environment of windows XP, a content-based image retrieval system is developed based on the proposed feature extraction algorithm of SIFT and rotation invariant LBP combination with Visual C++6.0and SQL2000database technology. Experiment results show that the SIFT and rotation invariant LBP combination algorithm has a good scale, rotation and viewpoint invariance. And it has a good robustness to illumination changes of images and has a higher retrieval accuracy than SIFT. As well, the designed content-based image retrieval system can meet the real-time retrieval requirement.