陳建中 Chien-Chung Chen

探討視覺神經機制 解釋整合影像資訊的知覺歷程

The major function of the human visual system is to extract useful information from the environment. For this purpose, the visual system has to integrate individual image elements into meaningful objects. However, under normal viewing condition, a visual scene can easily contain millions of image elements and thousands of objects.

Mathematically, one image element can be combined with any other image elements. Thus, there can be infinite number of possible interpretations for each scene. Yet, our visual system can easily form a stable and unique percept of each scene. My research is to understand h ow our visual system can achieve a coherent percept given what the visual system receives from an image are nothing but local elements such line segments. We started with the lateral interactions among the visual cortical neurons.

We documented how the spatial configuration of image elements can affect the interactions between visual cortical neurons tuned to different visual properties. This information allows us to predict what pairs of image elements are easily combined by the visual system than others. Also, based on these empirical results, we developed a computation model that can characterize the response properties of the visual cortical neurons affected by lateral interactions.

Once the interactions between two neighboring neural mechanisms are understood, we then explored the neural mechanisms underlying more complicate phenomena, such as color induction, or the color of a region of an image be affected by other regions; filling-in, or the percept of a regions of an image being replaced by neighboring regions; symmetry, or how the visual system compared pairs of images elements across space to understand a part of an image is a reflection of the other part; and Glass patterns, or the visual system combined multiple pairs of local elements to form a coherent global form percept.

Our results not only help us to understand the interaction among neural mechanisms but also see applications in building better image quality assessment metrics and experimental aesthetic measurements.