A Signal Processing Approach to Modeling Visual Masking

当前最先进的图像处理算法，用于将图像作为信号处理；这些技术之所以成功，是因为先进的信号处理技术和基于信号处理的人类视觉系统(HVS)模型。然而，考虑到颜色、结构甚至对象内容，许多应用程序更适合使用更高级别的图像视图。在该较高级别的视图中不存在图像或HVS处理的统一模型；这样的模型可以实质上推进图像处理中的当前技术水平。这项研究涉及开发这样一个模型，并将其结合到实际算法中，如低码率压缩、允许识别的低码率人脸压缩和医学图像。人类视觉系统最初执行低水平信号分析，最终结束于认知。许多信号分析阶段的模型预测对简单刺激的反应，但这些模型对自然图像无效，因为人类对结构的高阶处理产生的认知效应不能仅用对简单刺激的反应来建模。认知被认为是在输入过程中逐渐发生的；人类视觉系统的任何一个部分都不是单独负责识别内部表征或将其转化为抽象概念的。从信号分析到认知的视觉处理的连续体表明，基于信号处理的人眼视觉系统建模可以扩展到包括一些结构性处理。这项研究通过三个心理物理实验开发了这样的模型。第一项是一项受控掩蔽研究，允许开发包含结构的视觉掩蔽模型。第二项研究对观察者接受扭曲的意愿进行评级，将接受程度与第一项研究的检测结果联系起来。第三项研究比较了使用结构表征和基于信号表征的图像的识别时间，量化了结构在认知任务中的使用。

Current state-of-the-art algorithms that process images for human usetreat images as signals; these techniques are successful because of both advanced signal processing techniques and signal-processing based human visual system (HVS) models. Many applications, however are better approached with a higher-level view of images, considering color, structure, or even object content. No unified model of either images or the HVS processing exists at this higher-level view; such a model can substantially advance the current state-of-the-art in image processing. This research involves developing such a model and incorporating it into practical algorithms such as low-rate compression, low-rate facial compression allowing recognition, and medical imaging.The human visual system initially performs low-level signal analysis and ultimately ends in cognition. Many models of the signal analysis stage which predict responses to simple stimuli, but these models fail for natural images because humans' higher-order processing of structure produces cognitive effects which cannot be modeled using only responses to simple stimuli. Cognition is thought to gradually occur during input processing; no single portion of the HVS is singularly responsible for recognition or transformation of the internal representation to an abstract concept. The continuum of visual processing from signal analysis to cognition suggests that signal-processing-based modeling of the HVS can be extended to include some structural processing. This research develops such models through three psychophysical experiments. The first is a controlled masking study allowing development of a visual masking model incorporating structure. The second rates observers' willingness to accept distortions, relating acceptance with detection results from the first study. The third compares recognition times for images represented using structural and signal-based representations, quantifying the use of structure in cognitive tasks.