Critical operations in low-level human vision

Citation

Watt R (1996) Critical operations in low-level human vision. International Journal of Imaging Systems and Technology, 7 (2), pp. 65-77. https://doi.org/10.1002/%28SICI%291098-1098%28199622%297%3A2%3C65%3A%3AAID-IMA1%3E3.0.CO%3B2-Q

Abstract
This article examines the mechanisms of the initial stages of human vision and describes the critical functions that these processes have to achieve. The basis for the report is a series of computational studies that have sought to place what is known about human vision in the contexts of real tasks. Most considerations of low-level vision have focused on filtering and pointwise nonlinearities. Extending beyond what is already known, the present studies have identified the need for specific processes, over and above what is already known, to deal with the problems of (1) Contrast equalization: Differences in local illumination due to shadows and highlights cause variations in mean luminance and also in local contrast-the amplitude of luminance variations. This has severe effects for the filtering stage. (2) Segmentation: The image itself is a continuous and uniform data structure. It needs to be split into small primitive features from which object representations can be built. (3) Grouping: The primitive features are almost bound to be too primitive to be useful, singly, for any real visual tasks. Related primitive features that belong together need to be grouped together in a relatively tight fashion. The nature of these difficulties is examined, and potential approaches are considered and illustrated. A major feature of the work described concerns the formal properties of the data structures that will be required to hold the different types of information. Three distinct structures are considered: images, image descriptions, and visual descriptions. The latter two of these are novel.

Journal
International Journal of Imaging Systems and Technology: Volume 7, Issue 2