An Infrared Stripe Noise Removal Method Based on Multi-Scale Wavelet Transform and Multinomial Sparse Representation

Abstract

The non-uniformity present in the infrared detector and readout circuit leads to significant stripe noises in the infrared images. The effect of these stripe noises on infrared images brings trouble to the subsequent research. The currently available algorithms for removing infrared streak noises cannot effectively protect the non-stripe information while removing the stripe noise. Compared with these algorithms, our algorithm uses a multi-scale wavelet transform to concentrate the streak noise by frequency into vertical components of different scale levels. Then, our algorithm analyzes the unique properties of the streak noise compared to the ideal vertical component. The denoising model of the vertical component at each level is established with its multinomial sparsity, and the streak noise is removed by the alternating direction method of multipliers (ADMM) algorithm for optimal calculation. To prove the usefulness of our algorithm, we carried out a large series of real experiments, comparing it with the most advanced algorithms in terms of both subjective determination and objective indices. The experimental results fully demonstrate the superiority and effectiveness of our algorithm.

An Innovative Approach for Removing Stripe Noise in Infrared Images

“…According to the researchers’ in-depth exploration, stripe noise has global sparsity and gradient sparsity along the stripe direction. Li Mingxuan used the L1 norm to sparsely represent stripe noise, forming a regularization term in the energy function, and used a sparse representation of clean images across stripes as fidelity terms to separate and minimize stripe noise, combined with methods such as the wavelet transform, resulting in a good denoising ability [ 13 , 14 ].…”

Wavelet Channel Attention Module With A Fusion Network For Single Image Deraining

“…Consequently, the stereotypical use of inverse discrete wavelet transform (IDWT) degrades image reconstruction. Similarly, multi-scale processing based destriping methods commonly employ downsampling techniques to achieve large receptive fields, such as pooling, stepwise convolution, and wavelet decomposition [21], [25], [29]. However, using these samplers often produce varying degrees of semantic and structural information loss, leading to ineffective feature representations.…”