Scale, translation and rotation invariant Wavelet Local Feature Descriptor

Abstract In this paper is presented a novel scale, translation and rotation (STR)-invariant 1D-descriptor, named Wavelet Local Feature Descriptor (WLFD). The methodology to construct the WLFD locates in three different scale pyramids keypoints to extract representative features of the image. These scale pyramids are built using the 2D multi-resolution representation by Haar wavelet. Hence, the translation invariance was achieved using keypoints, and the scale invariance was obtained via the scale pyramids. The rotation invariance is worked out by adding the intensity values of the given image filtered by a binary mask with a circumference of radius r centered on the keypoint; the result was stored in the descriptor. For each keypoint, forty binary masks were used ( r = 1 , 2 , … , 40 pixels); thus the STR-invariant 1D-WLFD is a vector of forty entries. The robustness of the STR-invariant 1D-WLFD was tested using images with different scales and rotations and comparing the results with the Scale Invariant Feature Transform (SIFT) and Speeded-Up Robust Features (SURF) descriptors. The WLFD shows results similar to SIFT and much higher than SURF. Another advantage that WLFD shows is being faster than SIFT, building the descriptors up to four times faster. Therefore, WLFD is a fast, efficient and easy-to-implement methodology.