Methods for local phase quantization in blur-insensitive image analysis

Image quality is often degraded by blur caused by, for example, misfocused optics or camera motion. Blurring may also deteriorate the performance of computer vision algorithms if the image features computed are sensitive to these degradations. In this paper, we present an image descriptor based on local phase quantization that is robust to centrally symmetric blur. The descriptor referred to as local phase quantization (LPQ) can be used to characterize the underlying image texture. We also present a decorrelation scheme and propose three approaches for extracting the local phase information. Different combinations of them result in totally six variants of the operator that can be used alternatively. We show experimentally that these operators have slightly varying performance under different blurring conditions. In all test cases, including also sharp images, the new descriptors can outperform two state-of-the-art methods, namely, local binary pattern (LBP) and a method based on Gabor filter banks.