计算机生专业英语设计讲述.ppt

Joint Cascade Face Detection and Alignment;Abstract;Introduction;Alignment Helps Detection: a Post Classifier We used the Viola-Jones detector in OpenCV with alow threshold to ensure a high recall. The detector output a lot of image windows, of which many are false positives. We split all the images into two parts and used the positive and negative output windows in the first part to train a linear SVM classifier, and tested all the output windows in the second part. We call it a post classifier as it operates on the output of a cascade detector. All windows are resized into 96*96 pixels. We compare three types of features, without and with alignment. 1. we divide the window into 6*6 non-overlapping cells and extract a SIFT descriptor in each cell. 2. we use a fixed mean face shape with 27 facial points and extract a SIFT descriptor centered on each point. 3. we align the 27 facial points using the alignment algorithm in [21] and extract a SIFT descriptor centered on each point. ([21]:Face Alignment at 3000 FPS via Regressing); It clearly shows that features at the aligned facial points are much more effective to separate those difficult windows.While effective, the face alignment based post classifier is brute force and too slow for a standard cascade detector when a high recall is desired. In our experiment, a low threshold is set to achieve 99% recall of the OpenCV detector and each image outputs 3000 windows on average. Applying the post classifier for those windows takes a few seconds.; To better exploit the alignment information, we propose a unified framework for cascade face detection and alignment. Cascade Detection: Without loss of generality, the classification score in the cascade detection can be written as: (1) The cascade detection is very fast because most negative image windows are rejected after evaluating only a few weak classifiers. ;Cascade Alignment: Recent work proposes a pose regression framework that combines pose indexed fea