Comparative Analysis of Image Enhancement using Adaptive Threshold

In today's generation, as medical applications keep growing on an everyday basis, it is becoming increasingly necessary to develop different enhancement techniques to help gather precise information efficiently from biomedical images. They suffer from various degradations and some of the most common ones are poor quality due to additive noise, low contrast, etc. The basic motivation of this paper is to develop an image enhancement algorithm using adaptive thresholding in the frequency domain. Biomedical images are transformed into the frequency domain using DCT, DST and Hadamard Transform algorithms. DCT and DST algorithms are implemented for block sizes of 8x8, 16x16 and 32x32. While Hadamard Transform is implemented for a block size of 4x4. Setting a threshold that can completely remove all the noise is a time-intensive process. Thus, upon converting the image into the frequency domain, a statistical parameter, i.e standard deviation is calculated to fix up a threshold. For each value of the threshold, the results are compared using MSE and PSNR. The best MSE and PSNR values obtained were for DCT, i.e "around 120" and "around 27 dB" respectively.