Abstract: The images that are obtained from low illumination scenes have problems such as low overall brightness，poor contrast，and loss of fine details，which affect their performance in image enhancement applications.To improve the low illumination image quality，complete the image structure and make the texture details natural and clear，a low illumination image enhancement algorithm based on spatial transformation and adaptive gray correction is proposed.The adaptive compression multi-scale Retinex algorithm with gray correction is used to process the original image to obtain the balanced image，thus avoiding the phenomenon of the image being too bright or too dark when the traditional Retinex algorithm is used to process the image globally.The balanced image formed through the spatial transformation method is used to obtain the frequency domain smooth image and spatial domain sharp image，to improve the overall brightness and contrast of the image，respectively.Thus，the details of object edges in the image are retained.The original image，frequency domain smooth image，and spatial domain sharpened image are fused using the multi-focus fusion algorithm to obtain the final image.The experimental results show that the mean，Mean Square（MS），Information Entropy（IE），and Average Gradient（AG） of the proposed algorithm are improved by 1.63%，0.89%，0.17%，and 1.91%，respectively，when compared with SSR，CLAHE，MBYC，and other similar algorithms.This can effectively improve the brightness，clarity，and contrast of low illumination images，and enhance image edge information and texture details.

Key words: image enhancement, Retinex algorithm, spatial transformation, frequency domain image, spatial domain image, multi-focus fusion

摘要： 在低照度场景下采集的图像存在整体亮度偏低、对比度较差、细节信息丢失等问题，影响其在图像增强应用领域中的性能。为提高低照度成像质量，并使图像结构完整且纹理细节自然清晰，提出一种空间转换与自适应灰度校正的低照度图像增强算法。采用带有灰度校正的自适应压缩多尺度Retinex算法对原始图像进行处理，得到均衡化图像，避免在传统Retinex算法对图像进行全局处理时产生图像过亮或过暗的现象，通过空间转换方法处理获得的均衡化图像，分别得到频率域平滑图像和空间域锐化图像，以提高图像的整体亮度和对比度，从而保留图像中物体边缘的细节信息。在此基础上，采用多聚焦融合算法将原始图像、频率域平滑图像和空间域锐化图像进行融合，得到最终图像。实验结果表明，相比SSR、CLAHE、MBYC等算法，该算法的均值、方差、信息熵和平均梯度分别平均提升1.63%、0.89%、0.17%和1.91%，能有效提升低照度图像的亮度、清晰度和对比度，增强图像边缘信息和纹理细节信息。

关键词: 图像增强, Retinex算法, 空间转换, 频率域图像, 空间域图像, 多聚焦融合

CLC Number: 

• TP391