Object Feature Extraction for Image Retrieval Based on Quadtree Segmented Blocks

This study proposed a new object feature extraction method that employs the quadtree decomposition. In the proposed method, we segmented the image into variable sized blocks, named homogeneous blocks, which are the units of feature extraction process. Because the quadtree decomposition can highlight the details of images, more feature information can be extracted from the visual important objects than from the monotone areas of the image. The experimental results show the image retrieval performance is effectively improved as compared with the pixel based method.