Abstract
This study extends the work of Kwon et al. (Vision Research, 2016) to investigate the respective contributions of boundary smoothness and region information, using fragmented closed curves embedded in noise. Boundary smoothness was manipulated by perturbing the orientations of the fragmented boundary segments by a jitter of ±20deg or by a random jitter. Region information was either present or absent, such that the color inside the target shape was either different from or similar to the color of region outside. Four conditions were tested: (i) jitter ±20deg with colored region, (ii) jitter ±20deg with no colored region, (iii) random jitter with colored region, and (iv) random jitter with no colored region. Subjects fixated inside the target shape. Performance, measured by d', was highest in condition (i) and lowest in (iv). An interaction was found such that the improvement from condition (iv) to (iii) was greater than the improvement from (ii) to (i). The effect of fixation was also verified in a control experiment, where performance dropped to chance level when the stimuli in condition (ii) were presented with fixation point outside of the target shape. A biologically-inspired computational model was developed to emulate human performance. The model uses a complex-log (aka log-polar) representation of the image, which is a good approximation of the mapping from the retina to the primary visual cortex. The model performs a graph-based shortest-path optimization with four parameters in the cost function: interpolation distance, turning angle in interpolation, consistency of color on the same side of the interpolating fragments, and color difference between the two sides of the contour fragment. The model was tested using the same stimuli as those used in the psychophysical experiment and its performance was similar to the performance of the subjects.