Feature-matching-based specific object segmentation

We describe an unsupervised specific object detection and segmentation method form matching features between images. We focus on extracting some features of images and choose useful features of specific object. It is first to match features of each two images in our training images for capturing some specific object´s features. Then we get important features using matching features with features of training image specific object. Finally, we segment testing image into many segmentations and choose those segmentations which contain some specific object´s features. Our approach is evaluated on the Caltech-101 database, shows that some segmentation results of testing images, such as face segmentation. Experimental results indicate that our method is effective for segmenting specific objects without any supervision.