Vibratory image feature extraction based on local log-polar symmetry

We propose a feature extraction method for a newly developed vibratory image sensor. The proposed method extracts local log-polar symmetry as an image feature. In this article, it is shown that log-polar symmetric patterns include well-known characteristic points such as edges, corners, peaks, etc. The principle of our method is as follows: periodic vibration of an image sensor modulates radial and angular components of the local image pattern into separated temporal frequency components of incident light on each pixel. Comb type filters after a photodetector decompose the radial and angular components, and accumulate their power in a frame time. These powers are read out for calculating unevenness between them. The unevenness indicates the strength of symmetry of the corresponding local image pattern. Because our method modulates the local image pattern using image sensor vibration, the extracted features are free from spatial quantization errors.