Automated Enhancement of Grayscale Images using a Fast and Scalable K-Means approach

Abstract

Image enhancement is a widely used technique in disciplines ranging from Medicine to Meteorology. A grayscale image can be regarded as a dataset in which each pixel has a spatial location and an intensity level. Enhancement of grayscale images is distinctly challenging, because of the inability of the human eye to distinguish between any two consecutive gray intensities, and due to the limited range of intensity levels among which the pixels of the image have to be redistributed, so as to make the whole image appear sharper than before. In this approach, initially, I alter some extreme intensities present in the image, and divide the residual ones into groups, known as clusters, based on their Eucledian distances from the clusters' mean intensities, using the K-Means clustering algorithm. Then, I change the final cluster means in proportion to their mutual distances, such that, the new means are spread over the entire range of possible intensity levels. Finally, I assign the pixels of the image to the intensity levels corresponding to the new means of the clusters to which the pixels' intensities before clustering were assigned. This technique was tested on satellite images of Kolkata and Mumbai, and on some other photographs. The clarity enhancement obtained was far better than what could be achieved by using conventional contrast enhancement tools like Contrast stretching and Histogram equalization. Moreover, this technique could process the 512 pixels X 512 pixels grayscale satellite images typically within 600 milliseconds, even on a slow computer running Java version 1.5 on a 568 MHz Intel Celeron processor. This fact clearly reflects its speed and scalability.