A rotationally invariant descriptor based on mixed intensity feature histograms

Highlights：

• Most descriptors usually encode a single local feature for each sample point in the image patch. To improve the robustness of descriptor, we propose a new method to encode more than one different local features for each sampling point and calculate 2D histogram for it.

• Unlike most existing descriptors based on intensity order information, we propose a novel framework to build local rotation invariant descriptor. Our proposed descriptor MIFH (Mixed Intensity Feature Histograms) is built from the 2D mixed intensity feature histogram. The experimental results show that our descriptor obtains better performance than other descriptors.

摘要

•The LIOP (Local Intensity Order Pattern) operator sorts the intensity of neighboring points around one sample point and encodes an intensity order information into this sampling point. In the experiments, we found that the relative order information of the neighboring points had better robustness than LIOP. Therefore, we propose a novel operator based on intensity of pixels, namely LIROP (Local Intensity Relative Order Pattern).•Most descriptors usually encode a single local feature for each sample point in the image patch. To improve the robustness of descriptor, we propose a new method to encode more than one different local features for each sampling point and calculate 2D histogram for it.•Unlike most existing descriptors based on intensity order information, we propose a novel framework to build local rotation invariant descriptor. Our proposed descriptor MIFH (Mixed Intensity Feature Histograms) is built from the 2D mixed intensity feature histogram. The experimental results show that our descriptor obtains better performance than other descriptors.