Multi-focus Image Fusion Using Reorganized DTT Moments and Sparse Representation

Abstract

Multi-focus image fusion is a widely accepted solution to solve the problem of narrow depth-of-field suffered by optical lenses in an imaging system that fails to capture an all-in-one focused image. This article puts forward a focus fusion framework in the transform domain using moments derived from orthogonal Tchebichef polynomials. The input images are represented as 8 $\times$ 8 coefficient images using discrete Tchebichef basis functions. The coefficients across all the sub-images are reordered to resemble a decimated multiscale decomposition with directional subbands at three scales and an approximation band. Inter-scale spectral significance is used to fuse the mid-level detail coefficients whereas the finest details having the highest resolution employ the sparse-coding technique. The approximation subband also utilizes the same sparse dictionary to select the relevant features. After repositioning the fused coefficients back to their initial locations, inverse DTT is applied to generate the final fused image. Empirical results support the efficacy of the approach in multi-focus fusion in terms of perceptual clarity and quantitative metrics.