Sparse Component Analysis (SCA) in Random-valued and Salt and Pepper Noise Removal

TL;DR

This paper introduces a novel sparse component analysis method in the DCT domain for effective impulse noise removal, capable of detecting and correcting corrupted pixels, and enhances performance by combining with median filtering.

Contribution

The paper presents a new SCA-based approach for impulse noise removal using DCT sparsity, including a simplified version for salt-and-pepper noise and a hybrid method with median filtering.

Findings

Effective removal of impulse noise demonstrated in experiments.

The method accurately detects and reconstructs corrupted pixels.

Combining with median filter improves results at high noise levels.

Abstract

In this paper, we propose a new method for impulse noise removal from images. It uses the sparsity of images in the Discrete Cosine Transform (DCT) domain. The zeros in this domain give us the exact mathematical equation to reconstruct the pixels that are corrupted by random-value impulse noises. The proposed method can also detect and correct the corrupted pixels. Moreover, in a simpler case that salt and pepper noise is the brightest and darkest pixels in the image, we propose a simpler version of our method. In addition to the proposed method, we suggest a combination of the traditional median filter method with our method to yield better results when the percentage of the corrupted samples is high.