Effective techniques for digital image processing

“A picture is worth a thousand words”. Digital images have not only sent film images to history but also found various applications in many fields such as in photography, medicine, engineering and science. There has been continuous research on digital image processing, and many state-of-the-art methods have been introduced. This research aims at investigating important problems and developing effective algorithms in the areas of digital image processing and computer vision. The important problems studied in this thesis include image filtering, color constancy and depth estimation. Image filtering is the first problem studied in the thesis. The goal of image filtering is to remove noise and undesirable textures in an image. Noise may be presented in an image during image formation, storage and transmission. Removing noise with awareness of fine details and sharp edges is a challenging task. This thesis proposes two effective methods for edge-aware image filtering. The basic idea of the first method is to identify a pixel which is heavily corrupted by noise. This pixel will be eliminated from the filtering process applied to its neighboring pixels. In the second method, the filtering process is formulated as an optimization problem constrained by the local smoothness in the output image. Color constancy is the second problem studied in this thesis. Color constancy accounts for the ability of maintaining the correct color of the scene when light sources vary. In a typical digital image acquisition system, an image sensor receives energy reflected from objects of the scene. The spectrum of energy depends on the object and illumination. Extracting light energy contributed by illumination is a crucial task of digital image processing. This thesis proposes an efficient way to automatically correct the color of an image based on low level image features. Due to its simplicity, the proposed method can be implemented in devices with low computational power such as consumer digital cameras. Depth estimation is the third problem investigated this thesis. Since a digital image is a 3D scene projected to a 2D plane, it does not directly convey the geometric structure of the scene. In order to infer the depth information, two images of the same scene are used. This thesis provides a detailed study on depth estimation methods and proposes an algorithm based on the stereo vision approach.Submission note: A thesis submitted in total fulfilment of the requirements for the degree of Doctor of Philosophy to the Department of Engineering, College of Science, Health and Engineering, La Trobe University, Bundoora