Enhancement Of Low Resolution Mobile Phone Images For Improved Interpretation Of Visual Information

Abstract

The enhancement of digital images is necessary in many fields of application of images. It makes it easier for human viewers and machines to extract, interpret, and where necessary perform further processing on the information contained in the images. In this thesis, a spatial-spectral enhancement solution is proposed for degraded images captured using a cell phone camera. The degradations addressed here are those caused by factors such as poor resolution, fading, noise, shadows, poor illumination, geometrical distortions and ink smears which may lead to erroneous interpretation of the image information. A Nokia model 1680 cell phone camera was used to capture the images which were then loaded to a computer using the Nokia PC suite software version 7.1 and converted to 8-bit grey-scale images using MATLAB version 7.14. The grey scale images were then interpolated using a bicubic method to a size of 480x640 pixels. The interpolated images were enhanced by contrast stretching followed by single-scale retinex enhancement. Optimal global thresholding was then used to binarise the images and finally, morphological dilation was performed to yield enhanced images. For geometrically distorted document images, an additional procedure based on 2-D processing was carried out on the morphologically dilated images to rectify the distortions. Computer simulation based experiments have been used to demonstrate the effectiveness of the proposed algorithm. The results show a significant improvement over those obtained using other proposed methods in terms quality measures and performance of optical character recognition applications.