Noise reduction from grayscale images

An image is a useful form of data that also possess visual appeal. It can be used to observe various processes as well as study certain phenomena. The applications extend from taking good quality images to more crucial applications like medical imaging. A good quality image provides ease in deriving accurate conclusions as the data available would offer high integrity but in a practical situation, images are often corrupted by noise due to various reasons as it is captured or transmitted over a channel. In this paper, we have used a decision tree-based detection method (DTBDM) to detect if the target pixel consists of noise or not. We have worked on the following three test images of MATLAB, namely `Boat', `Peppers' and `Airplane'. After scrutinizing the pixel, we have used different reconstruction filters, namely Median filter, Geometric Mean (GM) filter, Arithmetic Mean (AM) filter, and Harmonic Mean (HM) filter to restore the detected noisy pixels. We have proposed the best reconstruction filter for this detection method by analyzing the peak signal-tonoise ratio (PSNR) values after every restoration and their corresponding histograms.