120 likes | 458 Views
Context-aware Exposure Auto-correction. Global exposure auto-correction. over-exposed. under-exposed. low-contrast. input. automatic histogram stretching. Global exposure auto-correction. Detection: valid histogram range < threshold Method: stretch histogram, adjust gamma curve.
E N D
Global exposure auto-correction over-exposed under-exposed low-contrast input automatic histogram stretching
Global exposure auto-correction • Detection: valid histogram range < threshold • Method: stretch histogram, adjust gamma curve #: Globally over-exposed, under-exposed & low-contrast images • Test Images include party, family, vacation, landscape, street view, pets
Local exposure auto-correction • High dynamic range scene input Auto adjustment [WLPG] Local shadow / Highlight [ours]
Local exposure auto-correction • Back-lighting object input Auto adjustment [WLPG] Local shadow / Highlight [ours]
High dynamic range scene detection segment scene region sky region input extract features sky detection , , local contrast in scene region sky histogram scene histogram classifier confidence map of sky
Samples of high dynamic range scene • False: • True:
High dynamic range scene #: True HDR scene images / Test Images
Back-lighting object detection • The most attractive backlit object is human! extract features face detection Histogram, local contrast in face/body region input classifier Histogram of image body detection input
Samples of back-lighting object • False: • True:
Back-lighting object #: True backlit human images / Test Images
Summary • Global incorrect exposure v.s. local incorrect exposure • The “detected Human + Sky images” account for almost 66.5% of the whole test images