480 likes | 622 Vues
This comprehensive overview explores Gestalt principles to elucidate how we perceive images as global figures rather than isolated local events. Key principles like proximity, similarity, and good continuation guide the grouping of local elements into significant figures, aiding in boundary detection. We discuss real-world applications, the mechanisms of segmentation based on texture properties, and the role of visual systems in resolving ambiguities in contour perception. Techniques for detecting meaningful contours through various cues are also examined.
E N D
Segmentation and Grouping • Gestalt approach • Problem - We don’t perceive local events in an image - we perceive more global figures • Elucidate principles which determine grouping of local “things” in an image into figures • Proximity • Similarity • Pragnanz • good continuation • symmetry
Real world problems to which we can apply gestalt principles • Segmentation • determining where objects are in an image and what their boundaries are. • Grouping • grouping together stuff as part of the same object; for example, across occluders.
Boundary detection(Local) • Luminance edges
Boundary detection (Local) • Luminance edges • Use center-surround receptive fields
Finding meaningful contours in image • Local edge detection • Problems - false targets, misses
Finding meaningful contours in image • Local edge detection • Problems - false targets, misses • Solution 1: use other cues • Texture • Motion • Disparity
Boundary detection (Local) • Luminance edges • Use center-surround receptive fields • Texture edges
Boundary detection (Local) • Luminance edges • Use center-surround receptive fields • Texture edges • Sharp changes in orientation, scale of textures
Boundary detection (Local) • Luminance edges • Use center-surround receptive fields • Texture edges • Sharp changes in orientation, scale of textures • Disparity edges Left eye Right eye
Boundary detection (Local) • Luminance edges • Use center-surround receptive fields • Texture edges • Sharp changes in orientation, scale of textures • Disparity edges 000000000000000000000 000000000000000000000 000000222222222000000 000000222222222000000 000000222222222000000 000000222222222000000 000000000000000000000 000000000000000000000 = - Left eye Right eye
Boundary detection (Local) • Luminance edges • Use center-surround receptive fields • Texture edges • Sharp changes in orientation, scale of textures • Disparity edges • Motion edges
Boundary detection (Local) • Luminance edges • Use center-surround receptive fields • Texture edges • Sharp changes in orientation, scale of textures • Disparity edges • Motion edges
Paradigm • Look for textures which “pop-out” to observers. • Characterize texture properties which support texture pop-out - fill in the blank: • A figure pops-out from the background if its __________ (property of texture) differs from that of the background. • Logic: • Pop-out is the result of automatic, low-level segmentation processes.
Texture properties which the visual system uses to do segmentation • Brightness • Contrast • Scale • Orientation
Real-world Justification for these properties • Most objects in a scene will differ in at least one, and probably more of these properties. • When an object’s texture doesn’t differ from that of it’s background it is camouflaged. • But why only these and not others?
Mechanisms for texture segmentation • Texture is a semi-local property of an image • Texture is the “micro-pattern” in an image • An individual point in an image cannot have texture, but a small region can. • Complex cells are good coders of texture properties • have local receptive fields, but aren’t sensitive to position of a pattern within the receptive fields • Signals how much oriented “stuff” falls within their receptive fields
Cortical images • Treat a set of cortical cells with the same receptive field properties as an image. The activity of the cell whose receptive field is centered at a given position of the visual field is the “intensity” of the cortical image. • Have cortical images for each combination of orientation and scale preferences.
Complex cell images • A cortical image made by looking at the firing rates of complex cells with the same orientation and scale preferences. • Example: • Fine-scale, vertical complex cell image - • firing rates of complex cells with small, vertical receptive fields. • An image of the fine-scale vertical “stuff” in an image.
Problem - • How does visual system resolve ambiguities in local measures of image intensity changes to decide what is part of a contour and what isn’t? • How does the visual system integrate local edge information into global figures? • Phenomenal window into visual processing of contours: • illusory contours and amodal completion
Illusory contours and amodal completion are flip sides of the same coin • Amodal completion - Filling in boundaries of objects behind occluders. • Illusory contours - Filling in boundaries OF occluders. • The appearance of illusory contours usually coincides with the appearance of amodally completed boundaries.