A DCT-based multiscaled binary descriptor invariant to complex brightness changes

Binary descriptors have been very popular in recent years due to their computation and memory efficient capabilities. Many of them could deal with geometrical variations and linear intensity shifts, however, they were not examined for more complex brightness changes specifically, which might result of some internal conditions such as nonlinear camera capturing parameters or external conditions such as location of the light source, viewing angle, scene properties. Moreover, the existing binary descriptors are sensitive to photometric distortions such as noise and blur. In this study, our aim is developing a binary descriptor which is highly robust to complex brightness changes and photometric distortions. The preliminary results demonstrate the noticeable performance of the proposed method when the mentioned distortions were artificially added on testing images of an object recognition dataset. Keywords-binary descriptor; image understanding; memoryefficient; robust to photometric distortions