A multi scale guided filter for illumination estimation in retinex algorithm

3rd China-Romania Science and Technology Seminar

自然科学;工业技术

Guo, L.^1 ; An, D.^2,3 ; Xu, Y.^3 ; Li, H.^1 ; Shao, M.^3

School of Mechanical Engineering, Shenyang Jianzhu University, Shenyang

110168, China^1

Research Centre for Analysis and Detection Technology, Shenyang Jianzhu University, Shenyang

110168, China^2

Research Institute for Micronano Detection and Motion Control, Shenyang Jianzhu University, Shenyang

110168, China^3

关键词: Bi-dimensional empirical mode decompositions;    Illumination components;    Illumination estimation;    Information entropy;    Information loss;    Intrinsic Mode functions;    Reflection components;    Retinex algorithms;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/399/1/012022/pdf
DOI  :  10.1088/1757-899X/399/1/012022

来源: IOP

【 摘 要 】

The method of illumination estimation for images affects the effect of retinex algorithm and causes information loss and halo artefact. A multi scale guided filter is proposed to estimate the illumination. In this method, bidimensional empirical mode decomposition is combined with guided filter. Firstly, the image is decomposed into multiple bidimensional intrinsic mode functions and a trend function by bidimensional empirical mode decomposition. Secondly, we use guided filter to estimate illumination information of each bidimensional intrinsic mode functions. Then illuminance component is formed by combining the illuminance information and the trend function. Finally, guided filter is again used to achieve a more accurate illuminance component. The simulation results show that multi scale guided filter can estimate the illumination component deeply and ensure the integrity of the reflection component. In the effect image of retinex algorithm, the contrast is raised by 1, the luminance is raised by 1, the information entropy is raised by 0.18, the clarity is raised by 0.001, and the visual effect is improved as a whole.