Visual object segregation is a challenging task for an autonomous robot situated in the real world. The task involves two problems that depend on each other: what to find and where to find it. Because of this mutual dependence, the acquisition process of the visual segregation is more difficult. In this report, inspired by developmental vision researche, we propose a scheme to bootstrap the acquisition process initiated by the behavior-based visual segregation. To show the potential of the scheme, a prototype system was constructed on a miniature car with a video camera. During the voluntary wandering, the car associated optical flow field with the behavior and then succeeded in detecting anomalous regions from within the field. Localized anomalous regions were clipped and appeared very easy in finding specific features of the patterns. Such features might be useful for embodied representation of the territory where the robot behaves boldly.