Novel Image Recognition Based on Subspace and SIFT

Abstract

In the light of the deep analyses of subspace recognition and SIFT recognition, a novel image recognition based on subspace and SIFT is proposed to provide a recognition from global features to minutiae features. First, subspace is used to implement coarse image recognition, gaining one or more candidate samples with different identities. Then, a special SIFT recognition environment is designed, in which the approach takes all the images as objects, builds a multi-object sample image with its size limited below a certain size, detects SIFT points based on object regions and recognizes the test image through SIFT point registration statistical vote. The experiments show that the designed SIFT recognition environment can increase SIFT recognition accuracy and the method based on subspace and SIFT can provide accurate recognitions in a mass of images. Under some special environments, recognition accuracy tends to 100%.