类Haar特征(Haar-like features)完全讲解.pdfVIP

Haar-like Features and Integral Image Representation Master in Computer Vision and Artificial Intelligence 18th December 2009 David Gerónimo dgeronimo@cvc.uab.es www.cvc.uab.es/adas Index Haar features •Definition •Inspiration •Normalization •Different sets Problem •Computational cost •Definition Integral image representation Further tricks and extensions •Integral image for 45º rotated features •Integral image for almost any angle feature •Integral volume •Commutativity property Applications Haar features Computer Vision in the late 90s… Computationally expensive features: raw pixels, edges, etc. Haar features Computer Vision in the late 90s… Computationally expensive features: raw pixels, edges, etc. Inadequate image representations (features) to deal with complex objects: difficult to model with existing features. Haar features IEEE Conference on Computer Vision and Pattern Recognition 1997 Dr. C. Papageorgiou Dr. T. Poggio (et al.) Haar features Haar features, also known as Haar-like features, are a simple and inexpensive image features based on intensity differences between rectangle-based regions that share similar shapes to the Haar wavelets. FHaar E (Rwhite ) −E (Rblack ) R R white black [Oren97,Papageorgiou00] Haar features Haar features, also known as Haar-like features, are a simple and inexpensive image features based on intensity differences between rect