Abstract: PURPOSE: Object recognition involves the brain segregating objects from their surroundings. Neurophysiological studies of figure-ground texture segregation have yielded inconsistent results, particularly regarding whether V1 neurons are capable of figure-ground segregation, or simply detect texture borders. To address the issue, here we employed two-photon imaging to study V1/V4 population coding for figure-ground segregation in awake, fixating macaques.
Methods: We measured the neuronal responses for texture segregation in three V1 response FOVs, one from each awake, fixation macaques, and in six V4 response FOVs from another three awake, fixation macaques. The experimental texture stimuli were composed of oriented line segments. Two types of texture stimuli were mainly used: the uniform texture and the figure-ground texture. The uniform texture was a 32° × 32° patch composed of randomly positioned line segments with one of four orientations (0°, 45°, 90°, 135°). The figure-ground texture was composed of a 4° × 4° square figure texture superimposed on a 32° × 32° uniform ground with orthogonal orientations. During the recordings, the figure position varied relative to the pRF of the FOV, and V1 and V4 neuronal responses to the figure, figure-ground border, and ground during a passive-viewing task were recorded. To analyze the V1 and V4 population coding for texture segregation, we trained a three-stage linear support vector machine (SVM) to decode texture border and figure-ground information using PCA-transformed neuronal responses.
Results: When considering the average response changes, it appears that V1 neurons detect the figure-ground texture border instead of segregating the figure from the ground. Meanwhile, the role of V4 neurons in figure-ground segregation is uncertain due to their extremely small effect size. Our population coding results revealed that both V1 and V4 neurons can decode the texture border and segregate the figure from the ground with sufficient principal components (PCs). However, V1 neurons decoded figure-ground borders with considerably higher efficiency (requiring a few principal components) than V4 neurons. In contrast, V4 neurons were considerably more efficient than V1 neurons for figure-ground segregation.
Conclusions: These results indicate that V1 neurons are mainly responsible for border detection, and in addition provide rudimentary figure-ground information that is not very well represented by the first most informative PCs. However, V1 figure-ground information can be linearly read out and efficiently represented by downstream V4 neurons for segregation.

Key words: texture segregation, macaques, visual area V1 & V4, population coding, 2-photon calcium imaging