|M.Sc Student||Skarbnik Nikolay|
|Subject||The Importance of Phase in Image Processing|
|Department||Department of Electrical Engineering||Supervisor||Professor Emeritus Yehoshua Zeevi|
|Full Thesis text|
The phase of a signal is a non-trivial quantity. Usually it is ignored in favor of signal's magnitude. However, phase conveys more information regarding signal structure than magnitude does, especially in the case of images. It is therefore essential to use phase information in various signal/image processing schemes as well as in computer vision. This is true for global phase and, even more so, for local phase. The latter is sufficient for signal/image representation, while totally ignoring the magnitude information. The implementation of localized methods requires substantial computation resources. Thanks to the major progress in computing resources during the last decade, the implementation of localized methods has become feasible. Thus, there is a growing interest in localized approaches in both theory and application.
We first address the importance of phase with reference to several examples. We then present cases where local phase-based solutions are compared with magnitude-based ones. We argue that the phase has several advantages over magnitude: It is a unit-less bounded quantity, thus eliminating the need for thresholds. Local phase shows immunity to several types of noise, and variable conditions in signal strength (such as illumination in the case of images). We argue and demonstrate that due to the above-listed advantages, local phase is an important signal/image attribute to be used in signal/image analysis. Several novel local phase-based edge detection schemes are presented, discussed and compared with traditional edge detection schemes. We address the performance of our local phase-based schemes in the context of several applications such as: detection of edges, detection of man-made environment, image segmentation and denoising.
We demonstrate that local phase manipulations result in effects that are usually achieved by various methods: high/low pass filtering, box-filtering, derivative and even iterative diffusion schemes. Taking all of these together, this research implies that local phase has not been exploited in signal/image processing to its fullest extent so far and further research has the potential for additional results.
Our analysis and results indicate beyond any doubt that the use of phase is crucial for image processing.