Vision in V1 for Pre-attentive Segmentation
This paper analyzes computational strategies for segmentation, and argues that an initial stage of human visual segmentation, corresponding to pop-out, is a pre-attentive process we call segmentation without classification, i.e., separating an image into regions without classifying and comparing features within and between regions. This process has been proposed to be implemented in the primary visual cortex, or V1, which detects global region boundaries by detecting the breakdown of homogeneity or translation invariance in inputs. Local intra-cortical interactions mediate contextual influences on cells responses from outside their classical receptive fields, making neural activities near region boundaries higher than elsewhere, thus making boundaries more salient and pop-out, A biologically based model of V1 is used to implement the proposal. Examples of how the model performs texture segmentation and figure-ground segregation are demonstrated, thus linking seemingly unrelated data of V1 physiology/anatomy and visual psychophysics. It is further proposed that outputs from V1 provide a saliency map.
Zhaoping Li
University College London,London, WC1, U.K.
国际会议
8th International Conference on Neural Information Processing(ICONIP 2001)(第八届国际神经信息处理大会)
上海
英文
1069-1074
2001-11-14（万方平台首次上网日期，不代表论文的发表时间）