الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 141 |  |  |
| | | from | 141 | | |
Abstract
Computer-based technologies and tolls for industrial machine vision and scientific image processing are quickly being oncorporated into a variety of applications. The goal of machine vision is to process images acquired by cameras in order to produce an appropriate representation of the objects in the real world. The goal of this thesis is to create a computerized method for automatic face detection in cluttered scenes. The method we used is general and can be applied to many object detection problems with little modifications.
In this thesis, we present a neural network-based upright frontal face detection system. Generally, object detection is the problem of determining whether or not a sub-window of an image belongs to the set of images of an object of interest. The system task is to detect and locate upright frontal human faces. In a grayscale image that conatains human faces against cluttered background. The system proposes some solutions to the problems related to the face detection domain. It arbitrates between multiple neural networks and heuristics. such as the fact that faces rarely overlap in images. To improve the performance and accuracy of the used algorithm. We used a new preliminary algorithm to increase the speed of the system to 3-10 times faster than other systems. This new algorithm extracts all the 20*20 pixels sub-windows that have high possibility of containing faces, and process tham by the system.
The system as a whole is composed of the following teo stages: the preprocessing stage in which the system applies a set of filters to the image to increase the detection speed and reduce the variation caused by lighting or camera differences. The the nueral network detection stage , which is composed of three detection neural networks. The system arbitrates between the output of these three networks to get the best performance.
We used different arbitration and heuristics to implemant different systems. These systems were tested on 77 different test images containing a total 277 faces. These systems were able to detect betweem 83.1% and 94.6% of faces in all images with an acceptavle number of false detections. the proper system to be used depends on the application that this system will be used in.