1、An Adaptive Image Enhancement Algorithm for Face Detection,By Lizuo Jin, Shinichi Satoh, and Masao Sakauchi. ECE 738 In Young Chung,Outline,Motivation Problem in face detection Suggestion Basic idea of suggestion Approach Adaptive Image Enhancement AlgorithmResult and Comparison Conclusion,Problem i
2、n Face Detection?,Almost every face-detection methods depends on the intensity values of image Face detection under unconstrained condition result in failure because of the drastic variation of pixel intensity in face regions Image enhancement by intensity transformation can reduce this problem, wit
3、h histogram equalization (HE). HE applied to images with faces on a very light background, it may produce very dark face regions face detection failure.,Suggestion of solution,Solution?An adaptive image enhancement algorithm which is adapt to the intensity distribution within an image.,Basic idea,1.
4、 Why dont we make it even?,Entropy of darker pixels = Entropy of lighter pixels,2. Face is made up of many pixels,Face pixels make a cluster in histogram We can histogram ridge analysis technique,Approaches I.,EER (Entropy Error Rate) as an information theoretic measurerepresents the tendency of the
5、 information distribution within an imageIf EER is positive and largethe information lies mainly in the darker pixels If EER is negative large the information is lies mainly in the lighter pixels Goal : minimizing the EER,Where, S : estimate the relative position of the mean in histogramHD,HB : info
6、rmation in darker pixels and lighter pixels respectivelyHD,HB : the average entropy in either side,Approaches II.,Histogram Ridges Analysis : suggested in the paper “A fast histogram-clustering approach for multi-level thresholding” by Du-Ming Tsau and Ting-Hsiuing Chen Important parameter: the dist
7、ance between the leftmost and rightmost ridge because this distance is related with the intensity range of valid content in the image.,Adaptive Enhancement Algorithm,Step 1.,Extract Intensity Value in the input image,Step 2.,Compute the intensity histogram of the input image,Step 3.,Threshold the in
8、tensity histogram Against noise and stretch to 0,255,Smoothing with Gaussian smoothing Operator with variance = 2.0,Find valid ridges and distance between the ridges (Dr)this is related with the intensity range of valid content in the image.,Step 4.,Filter the histogram obtained in step 2 with a fil
9、tering coefficient to eliminate noise or unimportant details,Step 5.,Compress the detail region and expend important region by using entropy in darker and lighter side,Step 6.,Minimum EER finding process,After gamma correction with the parameter obtained in minimum EER F.P,Results,Before, Histogram
10、Enhancement,After Adaptive Enhancement,Comparison I.,Classical histogram equalization (HE),Adaptive histogram enhancement (AE),Comparison II.,Original image,HE,AE,Image with very light back ground,Conclusion and future works,Image enhancement is very important technique for face detection, especially in the images acquired in unconstrained illumination conditionUnsuitable enhancement can increase detection-failure rateAE algorithm estimate the image quality base on EER and intensity histogram and select best transform It performs much better than classical HE method,Question ?,
