Optimal gabor filter design for effective classification of images

The classification of images is now widely used in a range of applications. This thesis presents a new method to classify images of coal combustion which combined with Gabor filter, Fisher ratio and RBF neural network. Coal is important primary energy, but now the world is commonly facing a difficult problem of technology and environment which results from coal combustion. When obtained coal combustion images, we should to estimate this process of combustion normal or abnormal (included low-temperature and high-temperature). Researches have been done to expose the efficiency of 2D Gabor filter in edge detection, texture analysis, and image enhancement. Here a bank of Gabor filters is designed with multi-scale and multi-orientation and a group of filtered images with multi-scale and multi-orientation are obtained. Subsequently, Fisher criterion is used for selecting which Gabor filter parameters should be chosen for a class separability. Fisher rule finds the best line that suitable for classification, according to which the sample data is projected from a high-dimensional space to a low-dimensional space.Master of Science (Computer Control and Automation