Object Detection Using Contrast Enhancement and Dynamic Noise Reduction

Edge detection is one of the most important steps a computer must perform to gain understanding of an object in a digital image either from disk or from video feed. Edge detection allows for the computer to describe the shape of the objects in an image and create a pixel boundary defining what is considered part of an object, and what is not. Cannys edge detection algorithm is one of the most robust and accurate of these edge detection algorithms. However, as with many algorithms in image processing, there are many cases where the algorithm does not perform as well as an application requires. This can be caused by many problems, many of which are beyond the control of the image analyst because the images were supplied with poor lighting, or from security cameras, or low contrast situations. Even steps like converting the image to grayscale can interfere with detection. In this thesis we will explore improvements to the algorithm by a dynamic system that will select a color channel to help deal with data loss issues and improve the contrast between the object and its background using partial histograms. Then we will use histogram equalization to greatly improve the contrast of the image and explore a progressive implementation of histogram equalization to reduce the noise and get good detection of the objects that an unmodified edge detector would have struggled with.