【作者】 张明；

【导师】 李永杰；

【作者基本信息】 电子科技大学 ， 生物医学工程， 2018， 硕士

【摘要】 相机需要颜色恒常算法的参与来校正所拍摄到的图像的色偏,从时间上来讲,颜色恒常算法整体上在从低复杂度的静态方法到复杂的基于学习的方法过渡。最近,通过简单的实现方式,基于回归学习的颜色恒常方法使用一个简单的校正矩阵作用于一些低复杂度的颜色恒常方法上就可以取得极具竞争力的光源估计效果。然而,大多数基于回归学习的颜色恒常方法,例如Correct Moment(CM)方法,在测试的过程中对所有的测试集图像都使用一个相同的校正矩阵。考虑到相机拍摄到的图像的颜色通常受很多因素的影响,例如场景中的光源颜色以及反射表面的颜色,对所有的测试图像都使用一个相同的校正矩阵就显得不够合理。本文第一部分的工作首先运用数学手段,分析了影响基于回归学习的颜色恒常算法性能的主要因素,然后设计出一种可以自动选取适宜的部分训练集图像去为每一幅测试图像学习一个最优的校正矩阵的方法。通过这种改进策略,原始的基于回归学习的颜色恒常方法(例如CM)在四个广泛使用的图像数据库中的测试结果均得到了显著的提升。尽管基于学习的方法比传统的静态方法光源估计结果更为准确,然而目前大多数的基于学习的方法都缺乏一定的泛化能力,比如在一种相机或者一种场景中学习所得无法适用于其它类型的相机或场景。最直接的解决思路是构建一个包含有所有场景和相机类型的图像库,但对于极端场景,比如水下环境这种受到噪声、模糊等其余因素干扰的复杂场景,构建一个可以计算标准光源的图像库用于基于学习方法的训练尤为困难,水下图像的颜色恒常问题也就难以有效解决。本文第二部分的工作受硬骨鱼视网膜的结构和功能特性的启发,提出了一种水下图像增强模型来解决水下摄影中常遇到的噪声、色偏、画质模糊等问题。确切地说,模拟象鼻鱼特殊的光感受器结构来实现去噪,模拟水平细胞和视锥细胞之间的反馈以及一种红色通道补偿机制来实现颜色恒常。此外,双极细胞典型的中心外周拮抗式感受野以及另一条从无长突细胞到网间细胞再到水平细胞的反馈用来去除模糊。神经节细胞的颜色单拮抗细胞用来增强和校正颜色。我们的模型使用全局统计信息来引导所涉及的每一个低通滤波器的设计。大量的水下图像测试结果表明我们提出的模型可以有效解决上述问题。更多还原

【Abstract】 Camera needs color constancy algorithms to correct the color bias of the captured images,and the color constancy methods are overall improving from static methods to learning-based methods.Recently,with very simple implementation,regression-based color constancy(CC)methods have obtained very competitive performance by applying a correction matrix to the results of some low level-based CC.However,most regressionbased methods,e.g.,Corrected Moment(CM),apply the same correction matrix to all the test images.Considering that a captured image color is usually determined by various factors(e.g.,illuminant and surface reflectance),it is obviously not reasonable enough to apply a same correction to different test images without considering the intrinsic difference among images.Our first work mathematically analyze the key factors that may influence the performance of regression-based CC,and then we design a principled rule to automatically select the suitable training images to learn an optimal correction matrix for each test image.With this principled strategy,the original regression-based CC(e.g.,CM)is clearly improved to obtain more competitive performance on four generally used benchmark datasets.Although learning-based methods generally perform better than static methods,most of them are still lack of generalization ability,e.g.,learning from one type or one scene condition may not be able to apply to another camera or scene.An common and direct solution is to construct a dataset with all of camera types and scene conditions.Unfortunately,for the extreme scenes like underwater,which influenced by other factors such as noise and blur,it is really difficult to construct a dataset with ground truth illuminant for the learning process of learning-based methods.The problem of underwater CC is thus hard to have an effective solution.Our second work propose an underwater image enhancement model inspired by the morphology and function of the teleost fish retina.We aim to solve the common problems of underwater photography raised by noise,blurring and color bias.Specifically,the unique photoreceptor structure of elephantnose fishes is simulated to realize noise reduction,and the feedback from color-sensitive horizontal cells to cones and a red channel compensation are used to correct the color bias.Besides,the characteristic of center-surround opponency of the bipolar cells and another feedback from amacrine cells to interplexiform cells then to horizontal cells serve for haze removal.The single-opponent ganglion cells are used for color enhancement and color correction.Our model utilizes the global statistics to guide the design of each low-level filter,which realizes the self adaption of the main model parameters.Overall,our model is inspired by the image processing procedure in fish retina,and the experiments on extensive underwater images validate the effectiveness of our model.更多还原