Abstract:

The Retinex model refers to human perception of lightness and color. The multi-scale Retinex(MSR) algorithm, which is high dynamic range, color constancy and high color fidelity, is widely used in low-light image enhancement. A novel multi-scale retinex method based on sub-band weighting fusion for image enhancement was proposed. First, hybrid intensity transfer function was used to have retinex outputs in different scales. Then the retinex outputs were decomposed into non-overlapping spectral sub-bands. Image enhancement was processed in each sub-band. Final, the resulted image was fused by weighting every point of each Retinex outputs. The experimental results show that the proposed algorithm effectively enhances the contrast of original image and achieve a visual-pleasing and colorvivid outputs.

Key words: MSR; sub-band decomposition; hybrid intensity transfer function; image fusion; image enhancement