区域自适应的非均匀光照图像增强

Uneven Illumination Image Enhancement Based on Region Adaptation

传统的图像增强方法对非均匀光照图像进行增强时,通常只考虑了亮度和对比度的提升,忽视了增强过程产生的噪声放大和图像不自然等问题。而当前基于深度学习的方法仍然存在局部退化的现象,且训练模型需要庞大的数据集,运算量大,对硬件要求也更高。针对上述问题,提出了一种基于Retinex的区域自适应增强算法。首先,检测图像边缘并利用双边滤波器得到保留边缘信息的照明图。然后,引入光照调整因子,将部分光照调整到反射分量上以获得细节图像。这种方法可以改善细节并提高图像自然性。最后,通过结合改进的对比度受限自适应直方图均衡与伽马校正来增强图像对比度。实验结果表明,该算法对非均匀图像增强具有较好的普适性和鲁棒性,在保持图像自然性的同时,未出现明显过增强、光晕伪影、细节丢失和色彩失真等退化现象,在主观评价及客观评价方面均优于其他对比算法。

Traditional image enhancement methods only take into account the enhancement of brightness and contrast,but ignore the noise amplification and unnaturalness during the enhancement process.Besides,the current deep learning methods still have the phenomenon of local degradation,and the training model requires huge data sets,mass computation and high-performance hardware.In view of this,we propose a region adaptive enhancement algorithm based on Retinex.First,the edge of the image is detected and a bilateral filter is applied to obtain an illumination map which retains edge information.Then,an illumination adjustment factor is introduced,so that part of the light is returned to the reflectance component to obtain a detailed image.This step focuses on restoring the naturalness and details of the image.Finally,an improved contrast limited adaptive histogram equalization algorithm is combined with Gamma correction to improve the image contrast.According to the experimental results,the proposed method has excellent universality and robustness for uneven image enhancement,and performs better in preserving the naturalness of the image with no obvious degradation such as excessive enhancement and color distortion.The algorithm outperforms other competitive methods in terms of subjective assessment and objective assessment.