People with poor vision often use a video reader to assist them in reading documents. Existing systems magnify image detail but have limited video processing capability. Image enhancement techniques such as histogram modification, high-pass median filtering, thresholding, and frequency domain filtering improve clarity and compensate for these shortcomings. A key problem, however, is to automate these techniques to maximize system ease of use. 

In this project, digital images were imported into Matlab. An algorithm was developed for classifying images by whether they contain text or a picture. If an image was classified text, then a median filter was applied to the image to reduce impulse noise. Next, the image's histogram was analyzed, and an appropriate threshold was computed and applied to the image; this resulted in a much more readable black and white image. Conversely, if an image contained a picture, then its histogram was modified by histogram specification. The resultant picture image contained much improved contrast. Although not included in the final automation design, frequency domain techniques could then be applied to the resultant picture image to reduce any high frequency noise.
 

• (No web site available)
• Return to 1997 Project List
• Return to main Senior Project page