shadow_detection_and_removal-harini.ppt

* * * * * * * * * * * * * * * * * * * * * * Model based techniques relay on models representing the a priory knowledge of the geometry of the scene, the objects, and the illumination. Property based techniques identify shadows by using features such as geometry, brightness or color of shadows. (from “Cast shadow segmentation using invariant color features”) * * * * * * * * * * * * * * * * Shadow removal based on density. Shadows are lit by direct and ambient lights. Umbra The blackest part of a shadow from which all light is cut off. Penumbra A partial shadow, between regions of complete shadow and complete illumination. Brightness at any point E = (k Id cosθ +Ie)Rd θ = angle b/w incident light and object’s surface normal Id = luminance of direct light Ie =luminance of ambient light Rd= reflectance of the object surface. k = attenuation factor. Global brightness is calculated and density of shadow is estimated from this. Segmentation of shadow is done using shadow density. Color and brightness adjustments are performed for shadow removal. Color average and variance of umbra are adjusted. Results Laplacian pyramid based shadow removal Restoration process Supports restoration of shadowed surfaces with texture contrast and noise level similar to original lit region. Conditions One primary light source Shadow has to be cast on to scene surface. 3 masks: Ms= area of surface with all shadowed pixels. Ml= area of same surface with all lit pixels. Mshadow = binary mask of all pixels where shadow removal is required. Shadow seed specified by the user. Illumination invariant color distance measure is used find pixels in same surface as shadow seed. For small distances ,pixels have similar color vectors. These are saved into mask A. Shadowed and lit regions are determined and saved into Ms and Ml. Algorithm used Initialisation of Ms as the shadow seed and Ml = A - Ms. In each iteration: Pixels from Ml that are spatially adjacent and have similar colors