针对目前图像增强算法对于电子内镜图像光照不均匀、低照度区域边缘细节不明显以及高噪声等问题的局限性, 设计了一种用于电子内镜图像的融合低噪声、均衡光照和细节增强的Retinex框架, 并根据此框架设计了增强算法。算法首先利用基于位置信息与相邻频率的滤波器得到低噪光照值; 为了有效区分噪声与细节信息, 设计了一种基于最大后验概率估计(Maximum A Posteriori estimation, MAP)的反射率估计方法, 引入光照因子控制概率权重, 对于低照度区域反射率平滑项施加强约束, 并通过最大化其后验概率以应对低照度区域的高噪声问题; 为均衡光照、应对人体内黏膜和消化液的散射和吸收导致的图像退化, 基于暗通道先验(Dark Channel Prior, DCP)算法设计了反向均衡化模型以得到均衡光照值; 为应对低照度区域细节信息不明显问题, 利用对比度限制自适应直方图均衡化得到细节增强结果。通过使用均衡光照值补偿增强后的反射率, 实现噪声抑制、光照均衡、细节增强提高之间的有效融合。实验结果表明, 本算法较于近期的同类算法NIEIE(Non-uniform Illumination Endoscopic Imaging Enhancement), 能够在保持信息熵与峰值信噪比的基础上, 增强度提升23.94%, 对于电子内镜图像具有良好的适用性。

In view of the limitation of current enhancement algorithms for the problems of nonuniform illumination, unobvious edge details, and image noise in electronic endoscope images, a Retinex-based framework for combining low noise, balanced illumination, and detail enhancement of electronic endoscope images was proposed, and an enhancement algorithm was designed according to this framework. The algorithm used an illumination filter to obtain the illumination estimation, with the noise removed. To effectively distinguish noise from detailed information to obtain a more accurate reflectance, first, a reflectance estimation method was designed based on Maximum A Posterior estimation (MAP). The illumination factor was designed according to the illumination estimation to control the probability weight. The smoothness term of reflectance in a low illumination area was subjected to a strong constraint, and the posterior probability was maximized to cope with noise interference caused by nonuniform illumination. Second, a reverse equalization model was designed according to the Dark Channel Prior (DCP) algorithm to deal with local image degradation caused by nonuniform illumination and by the scattering and absorption of mucosa and digestive juice in the human body. As reflectance containing detailed information pays more attention to contrast, the contrast enhancement result was derived by using contrast-limited adaptive histogram equalization to deal with unobvious edge details. The final enhanced image was obtained by compensating the adjusted illumination back to the reflectance. Through this synthesis, the enhanced image represented a compromise between noise filtering, detail enhancement, and local contrast enhancement. The experimental results show that, compared with the average enhancement degree of the algorithm of Non-uniform Illumination Endoscopic Imaging Enhancement (NIEIE), that of the algorithm proposed in this paper increase by 23.94% by maintaining the information entropy and peak signal-to-noise ratio. Therefore, the proposed algorithm has good applicability for electronic endoscope images.