Infrared and visible image fusion and denoising via ℓ2−ℓp norm minimization

Abstract Most traditional infrared and visible image fusion methods often ignore noise in acquisition or transmission and their performance inevitably decreases in practical applications. To address this problem, a new and effective variational model is proposed for simultaneous image fusion and denoising. In an l 2 − l p norm minimization setting with p = 0 and p = 1 respectively, the hybrid l2 norm fidelity term is built to preserve image intensity and details from both infrared and visible images. And the nonconvex l0 norm and convex l1 norm sparsity constraints are applied to reduce noise while preserving important image fine features. Furthermore, a computationally efficient numerical algorithm based on half-quadratic splitting iteration is used to solve the complex optimization problem. Experimental results demonstrate that the proposed method can achieve a superior performance compared with existing fusion methods in both subjective and objective assessments.