视觉图像辨认眼动中的Top－down信息处理  

TOP-DOWN INFORMATION PROCESSING IN COGNITION EYE MOVEMENTS

作　　者：孙复川[1] 赵信珍[1] 顾凡及[1] 

机构地区：[1]中国科学院上海生理研究所,复旦大学生理学和生物物理学系

出　　处：《生物物理学报》1994年第3期431-438,共8页Acta Biophysica Sinica

基　　金：国家自然科学基金;中国科学院生物物理所视觉信息处理开放实验室研究课题

摘　　要：在视觉图像辨认过程中，眼球不是均匀地扫描全幅图像，而是通过一系列快速的眼球跳动来改变注视点位置，有选择地通过注视停顿来采集图象中的关键信息。通过实验对不同图像刺激时的眼动轨迹进行记录与分析，发现：（１）对于简单的几何图形，眼动注视停顿主要集中在图像中几何特征之处，亦即与周围不同的奇异点上；（２）对复杂图象刺激，眼动注视点位置决定于受试者的已有概念模型及其兴趣所在；（３）对中文单字进行辩认时，其眼动模式也是取决于受试者对该单字的知识（也即概念模型）。以上结果提示，视觉图象辨认主要是通过自上而下（ｔｏｐ－ｄｏｗｎ）的信息处理方式才完成．由中枢控制眼球运动，将注视点落到中枢决定的图形奇点上来，通过注视停顿对中枢认为的关键信息之处进行抽提，以实现辨认。这种处理方式不是只取决于输入的图像信息，也不必对目标图像的每个象素进行处理，而只需对图象中少量的关键信息部位进行重点的检测和处理，从而提高了图象信息处理的能力及效率。During visual image cognition , the eye-movement pattern is not scanning uniformly throughout the entire image ,but only consists of saccades and local fixatin pauses.The rapid saccadic movements change the positions of the eye and the fixation pauses between saccades acquire key infonnation of the cage. hi present inveStigation, the eye-movement twbories un colded and analed while a variety of cage sthauh presented ti subjects. The experunebtak data have shown that: (1) for sable geometric patterns, the eye-movement fixation pauses concentrated on the special geometric features of the pattern,i.e.those singularities,of which the characteristics are different from their neighbor regions,(2)for complex pictures,the locations of fixations depended on subjects' conceptUal model and motivation,(3)for Chinere characters, the eye-movement patternS also depended on subjan'knOWledge and familiarity with the charaCters (i.e.the coneeptual model). These experimental adults indiCate that the visual cage coition is mainly a top-down process. For aeqchition of suffidri infonnationfrom the image efficiently,the cential nervous system(CNS)controls eye movements to fix on the points which are singularities of the geometric pattern or contain esential infonnation of the image.With top-down processing,it is efficient for image coghtion because the visual system does not have to precess each pixels of the target, but deal with the key regionsselected by CNS.

关 键 词：眼动 图象辨认 视觉 信息处理 

分 类 号：Q811.4[生物学—生物工程] R339.14[医药卫生—人体生理学]