我们有两只眼睛，但只“看”到一个世界；外部世界的物体包含多维度的信息，我们却只看到一个维度捆绑的物体。独立的双眼多维度信息如何进行整合从而形成对外部世界的单一表征是知觉、计算视觉与神经科学领域最引人入胜的根本问题之一(Meese, Georgeson, & Baker, 2006)。本研究在实验室以往阈上水平的对比度整合和相位双眼整合等工作的基础上继续研究阈值水平对比度检测任务的双眼整合特性并测量双眼整合特性与双眼竞争、立体视等双眼视觉现象间的关系。为此本研究设计了2个研究内容共包括4个实验：
研究一主要围绕阈值水平对比度信息的双眼整合特性展开，包括两个实验。实验一为阈值水平对比度的视觉反应特性研究，测量多种条件下的对比度反应特性。知觉模板模型的分析结果显示对比敏感度主要受内部加法噪音和模板增益的影响，其中内部加法噪音随空间频率改变的幅度更大。实验二用双眼分视及眼间掩蔽实验范式测量单双眼条件、不同外部噪音水平及空间频率下的对比敏感度，结果发现不同眼间条件（F (4, 8) = 5.06, p = 0.025）及外部噪音（F(2,4)=11.40，p=0.022）条件下的对比敏感度存在显著的差异，而四种单眼条件下的对比敏感度则没有差异（F (3, 159)=0.89, p=0.45）。为解析阈值水平的对比度整合特性，研究一构建了双眼知觉模板模型。新模型由单眼模板增益（β），单眼非线性转换函数（γ1），单眼增益控制前内部噪音（N1），单眼对比度增益控制（b），单眼除法过程，双眼整合阶段（γ2），双眼内部噪音（N2）及决策阶段等八个成份组成。3名观察者的最优简化模型平均起来可解释92%以上的数据变异量。
研究二主要围绕双眼竞争和立体视与双眼对比度、相位整合之间的关系展开。实验三测量了正常与异常眼间条件下的双眼整合及双眼竞争特性。模型分析结果表明非优势眼的单眼信号衰减在双眼整合中起主要作用；双眼竞争结果与模型参数的回归分析发现单眼信号衰减对两眼占优势总时长及总次数的比值的回归显著，与平均每次占优势时长之比间的回归为边缘显著。实验四测量了观察者在多种视差水平及眼间对比度水平下的立体视、对比度和相位双眼整合特性。结果发现观察者的相位整合和立体视辨别任务表现出对比度特异性，对比度的双眼整合则为相位/视差非特异性。多通道对比度增益控制模型分析的结果表明仅包括单眼信号衰减的模型为最优模型，这在一定程度上说明单眼信号衰减可能是影响双眼对比度、相位整合及立体视辨别的主要因素之一。
以上研究结果将可大大加深对双眼对比度信息整合基本规律的认识，可为理解双眼整合的生理意义和建构立体视的认知与计算模型提供重要借鉴。平衡的双眼输入不但是视觉系统正常发育的前提，也对阅读、手眼协调、去掩蔽等复杂任务至关重要，在明确双眼信息整合基本规律的基础上探索双眼信息整合调控的有效途径，也将为弱视等双眼信息整合异常疾病的机制和治疗提供新见解。

We have two eyes, but only “see” one world. The visual information we get from the outside world usually was consisted of many dimensions. How the independent information combines into one unique image to represent the outside world is one of the most fascinating questions in perception, computational science and neuroscience fields (Meese et al., 2006). This research is trying to systematically study the binocular combination characteristics for both threshold and supra-threshold levels and its relationship with the binocular phenomena, such as binocular rivalry and stereopsis. To fulfill this purpose, we designed two projects which contains four experiments:
Projects 1 mainly focuses on the threshold level’s binocular contrast information combine characteristics, which was composed of two experiments. Experiment 1 is about the visual system’s response characteristics for the signal contrast under threshold level, which measured the contrast response properties for many conditions. The perceptual template model analysis results showed that the internal additive noise and template gain were the mainly factors that affect the threshold level contrast sensitivity, of which the internal additive noise varied more significantly. Experiment 2 applied the dichoptic and interocular masking paradigm to measure the contrast sensitivity for monocular and binocular conditions under different external and spatial frequencies, significant difference was found for the contrast sensitivity under eye conditions(F (4, 8) = 5.06, p = 0.025) and external noise levels (F(2,4)=11.40，p=0.022)）, while no difference was found for the four monocular conditions (F (3, 159) =0.89, p=0.45). In order to illustrate the mutual interaction between two eyes during the contrast combination procedure, the binocular perceptual template model was developed which consisted of seven components: (1) monocular template gain (β), (2) the monocular non-linear transducer tranfer fucntion (γ1), (3) the pre-contrast-gain control internal noise (N1), (4) interocular contrast gain control, (5) binocular contrast combination stage (γ2), (6) binocular internal noise (N2) and decision stage. Modeling analysis results showed that the best fitted model was the one with only internal noise N3 (the summation of internal noise) changing with spaital frequencies, which would explain the data variance over 92%.
Project 2 concentrates on the relationship between binocular phenomena (binocular rivalry and stereopsis) and binocular contrast and phase combination characteristics. Experiment 3 measured the binocular combination and rivalry property under both normal and imnormal eye conditions. Modeling results showed that the monocular attenuation was the major decisive factor in the binocular contrast and phase combination procedure. Regression analysis showed that the ratio of the total dominate duration and total dominate times changed significantly with monocular attenuation, while the regression for the average dominate duration was marginal significant. Experiment 4 is about the normal observer’s binocular phase, contrast and stereopsis characteristics under multi-disparity and interocular contrast ratio levels. Results showed that both the phase combination and disparity threshold demonstrated the contrast dependent characteristics, while inversely the contrast combination was independent from phase or disparity levels. The multi-channel contrast gain control model analysis results showed that the model with monocular signal attenuation (and direct interocular inhibition) was the best fitted model, this may demonstrate that the monocular signal attenuation may be one of the major factors that influence binocular contrast, phase combination and stereopsis discrimination.
This research will strengthen our understanding of the binocular signal combination principles and provide important reference for the comprehension of binocular integration’s physiological meaning and the cognitive and computational model construction for stereopsis. Balanced binocular input is not only the premise of the vision system’s normal development, but also is important to some complex tasks, such as reading, hand eye coordination and so on. Based on defining the basic rules of binocular information integration, it is necessary to explore the effective ways of binocular integration regulations and will also provide new insights into the mechanism and treatment for abnormal visual systems’ binocular information integration, such as amblyopia.