Video based Handwritten Characters Recognition

Video as an input device1 is becoming a viable method for information products because of its lower cost along with increasing power of computer processor. Its supporting humancomputer communication is evident. (Mann, 1996) used video-cam as the input device of a wearable computer system has reported that video becomes a "visual memory prosthetic", a perception enhancer, allowing one to remember faces better, and see things that normally would never have been seen. Some researches in this area are of assistive technology oriented. (Betke et al., 2002) have developed a visual tracking system that interprets visiblelight video to provide computer access for people, in particular, children with severe disabilities. Their system “Camera Mouse” was named because it tracked a body feature like the tip of the nose with a video camera and used the detected motion of the feature to directly control the mouse pointer. However, the spelling board was used to control what was typed by selecting the letters with the Camera Mouse. It took more than 20 seconds to input a character. (Cantzler & Hoile, 2003) proposed a novel form of a pointing device to control the mouse pointer by moving the camera freely in space in a similar fashion to the way a laser pointer controls a laser dot. They locate the computer display in the camera image by finding structured or large regions. The chosen method exploits the fact that there is much structured content in screen. This is sometimes too restrictive for diversity of window applications. Another vision-based pointer device, Gray level VisualGlove, intended for wearable computers, was presented by (Iannizzotto et al., 2001). While previous methods are mainly based on color for image segmentation and human skin detection, their system only relies on gray level gradient to be of low computation cost. Template matching with the highest degree of confidence, the lowest Hausdorff distance, is chosen as a possible detected object. However, it could not deal with scaled or rotated finger motion.