Removal of Random Valued Impulsive Noise

Abstract

In digital Image Processing, removal of noise is a highly demanded area of research.Impulsive noise is common in images which arise at the time of image acquisition
and or transmission of images. Impulsive noise can be classified into two categories,namely Salt & Pepper Noise (SPN) and Random Valued Impulsive Noise (RVIN). Removal
SPN is easier as compared to RVIN due to its characteristics. The present work concentrates on removal of RVIN from images.Most of the nonlinear filters used in removal of impulsive noise work in two phases,i.e. detection followed by filtering only the corrupted pixels keeping uncorrupted ones intact. Performance of such filters is dependent on the performance of detection schemes. In this work, thrust has been put to devise an accurate detection scheme and a novel weighted median filtering mechanism.
The proposed detection scheme utilizes double difference among the pixels in a test window. The difference is computed along four directions namely, horizontal, vertical,and two diagonals to capture the edge direction if any exists. This helps to identify, whether the test pixels under consideration is an edge pixel or a noisy one. Subsequently, the corrupted pixels are passed through in weighted median filter, where more weights are assigned to those pixels which lie in a minimum variance direction among all the four. Extensive simulation has been carried out at various noise conditions and with different standard images. Comparative analysis has been made with existing standard schemes with suitable parameters such as Peak Signal to Noise Ratio (PSNR), fault detection and misses. It has been observed in general that the proposed schemes outperforms its counterparts at low and medium noise conditions and performs at par at high noise conditions with low computational overhead. The low computational requirements have been substantiated with number of operations required for single window
operation and overall time required for detection and filtering operation.
In addition, every detector utilizes a threshold value which is compared with a predefined computed value to decide whether the pixel under consideration is corrupted.
Fixed threshold may perform well for one image at a particular noise condition. However, generalization is not possible for a fixed threshold. Hence, requirement for an
adaptive threshold is realized. In the later part of this thesis, we propose an impulsive detection scheme using an adaptive threshold. The adaptive threshold is determined
from an Artificial Neural Network (ANN) using various statistical parameters of noisy image like (µ, σ2, µ4) as inputs. The performance of this scheme is also compared with simulation results.