Low Illumination Image Color Enhancement Based On Illumination And Reflection Analysis

The color of the objects is determined jointly by the illumination and the inherent reflection coefficient of the objects, however, the human eyes have the ability to remove the influence of illumination adaptively based on the human visual system. This visual characteristic, recognizing the real color of the objects in different external illumination environment, is called human visual color constancy. The image formation system of the image acquisition device imitates the human eye vision, however, machine vision lacks of the analysis process of the cerebral vision cortex and is incapable of removing the influence of illumination. Especially in low illumination environment, the image formation system can't record the real color of objects accurately, which severely affects the subjective quality of the color digital image. Therefore, how to suppress the influence of low illumination on the image colors effectively, and provide realistic and reliable color information for the subsequent image processing, has always been one of the key issu es in the research field of color digital image processing. The main research work of this thesis is to correct and enhance the low illumination color digital images.The conventional multi-scale center/surround Retinex color correction algorithms may lea d to noise amplification and pseudo halo phenomenon in processing the low illumination image. A multi-scale center/surround Retinex low illumination image color correction algorithm based on bilateral filtering is proposed in this thesis, which reduces the noise of the reflection coefficient variable using the illumination information and performs the unconventional gamma correction on the reflection coefficient variable. Firstly, according to the multi-scale center/surround Retinex color correction theory, the proposed algorithm divides the reflection coefficient image into different regions and filters each region of the reflection coefficient image by bilateral filter with the luminance distribution characteristics of the illumination information image. Secondly, the illumination information image and the reflection coefficient image corresponding to the original low illumination image are reconstructed in terms of brightness and contrast differently with the gamma correction thought, which enhances image brightness and highlights image detail under the condition that the noise is restrained to eliminate the phenomenon of pseudo halo. Finally, the correct illumination information image and the correct reflection coefficient image are merged into one to increase the low frequency information of the merged image, which makes the image more natural and smooth. The subjective quality of the low illumination color digital image can be improved efficiently.The core concept of the color correction algorithms based on the Retinex theory is reasonably estimating the illumination information of the original image and removing the influence of the illumination information. However, in the color enhancement for low illumination image, Retinex algorithms destroy and aba ndon the available information in the original low illumination image, which leads to the loss of the image details and the blurred texture. Reflection coefficient image expresses color more powerful than the original low illumination image and yet has weaker expression in image details. Therefore, a low illumination image color enhancement algorithm based on local color transfer theory is proposed in this thesis. Considering the different distribution of the color in original low illumination image, the proposed algorithm performs color segmentation on the reflection coefficient image using the Mean Shift theory in the space and color five domains, and transfers the color distribution characteristics from the reflection coefficient image to the original low illumination image locally according to the global and local characteristics of the color distribution in the reflection coefficient image. Compared to the traditional multi-scale center/surround Retinex algorithms, the proposed algorithm can eliminate th e color offset in low illumination image effectively and improve the saturation and contrast of the corrected image significantly.