An Iterative Approach to Improved Local Phase Coherence Estimation

This paper introduces a novel iterative approach to estimating local phase coherence in situations characterized by low signal-to-noise ratios. Local phase coherence is used for a wide range of computer vision applications such as edge and corner detection and object description. An issue faced in extracting local phase coherence is the presence of image noise. While existing approaches to dealing with noise when estimating local phase coherence is effective for low noise situations, they are inadequate for situations contaminated by high levels of noise. In the proposed approach, the issue of high image noise is addressed by re-estimating both the local phase coherence and the underlying image content iteratively to improve local phase coherence estimates. This is performed using a feedback loop model, where the local phase coherence estimates are used to re-estimate the image content using a moment-adaptive bilateral estimation scheme and information from the re-estimated image content is used to re-estimate the local phase coherence. Experiment results show that the proposed approach can be used to provide improved local phase coherence estimates.