B. Heisele, T. Poggio, and M. Pontil. Face detection in still gray images. A.I. Memo 1687, Massachusetts Institute of Technology, May 2000.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....likelihood and logistic regression. In text classification and information retrieval support vector machines [48] 161] and transductive support vector machines [100] surpassed the popular naive Bayes and generative text models. In computer vision, person detection recognition [148] 52] 142] [71] and gender classification have been dominated by SVM frameworks which surpass maximum likelihood generative models and approach human performance [138] In genomics and bioinformatics, discriminative systems play a crucial role [202] 208] 81] Furthermore, in speech recognition, discriminative ....

B. Heisel, T. Poggio, and M. Pontil. Face detection in still gray images. Technical Report 1687, Massachusetts Institute of Technology, 2000. AI Memo.

....popular, but as a general representation for human body parts it is unclear whether standard (rectangular) or non standard (square) wavelet constructions are most suitable. Heisele et al. recently obtained better results for their SVM face detector using gray levels rather than Haar wavelets[3]. Some authors also feel that wavelets are unsuitable as a general image representation because they represent point events rather than line or curve ones. As an alternative, they propose ridgelets and curvelets, which could prove useful for detecting human limbs. Here we leave such issues for ....

B. Heisele, T. Poggio, and M. Pontil. Face detection in still gray images. Technical report, AI Memo 1687, Massachusetts Institute of Technology, 2000., 2000.

....a single optimized VTU, but with identical setting for the earher hierarchical stages. The plot shows the combined rate of correctly identified faces over the rate of misclassifications as a fraction of all non face images for the hierarchy and a polynomial kernel support vector machine classifier [3]. 3 Results In Figure 2a we compare the classification performance of our model, using C2 activity nearest neighbor matching, to the results published in [12] using the SNoW recognition approach and applying a linear support vector machine on the COIL 100 dataset. To show the application to ....

....of unknown stimuli. With an identical setting as described above, however, using a single sigmoidal output VTU with s = tanh(r 2) we performed gradient based supervised optimization of r with target outputs of 0.9 and 0.9 for nonfaces and faces respectively. The training ensemble (data from [3]) consists of 2429 19 x 19 pixel face images and 4548 non face images. A threshold criterion was used to decide the presence or non presence of a face for a different test set of 472 faces and 23573 non faces. The non face images consist of a subset of all non face images that were found to be ....

[Article contains additional citation context not shown here]

B. Heisele, T. Poggio, and M. Pontil. Face detection in still gray images. Technical report, MIT A.I. Memo 1687, 2000.

No context found.

B. Heisele, T. Poggio, and M. Pontil. Face detection in still gray images. A.I. Memo / C.B.C.L Paper 1687, Center for Biological and Computational Learning, MIT, Cambridge, MA, 2000.

No context found.

T. Poggio B. Heisele and M Pontil. Face detection in still gray images. Technical Report A.I. Memo No. 1687, MIT, May, 2000.

No context found.

B. Heisele, T. Poggio, M. Pontil, Face detection in still gray images, AI Memo 1687, Center for Biological and Computational Learning, MIT, Cambridge, MA, 2000.

No context found.

B. Heisele, T. Poggio, and M. Pontil. Face detection in still gray images. A.I. memo 1687, Center for Biological and Computational Learning, MIT, Cambridge, MA, 2000.

No context found.

B. Heisele, T. Poggio, and M. Pontil. Face detection in still gray images. AI Memo 1687, Center for Biological and Computational Learning, MIT, Cambridge, MA, 2000.

....Both global systems described in this paper consist of a face detection stage where the face is detected and extracted from an input image and a recognition stage where the person s identity is established. 3.1. Face detection We developed a face detector similar to the one described in [8]. In order to detect faces at different scales we first computed a resolution pyramid for the input image and then shifted a ## # ## window over each image in the pyramid. We applied two preprocessing steps to the gray images to compensate for certain sources of image variations [19] A best fit ....

....experiments we divided the face images of a person into four clusters. This is not exactly a one vs all classifier since images of the same person but from different clusters were omitted. 4.1. Detection We implemented a two level component based face detector which is described in detail in [8]. The principles of the system are illustrated in Fig. 3. On the first level, component classifiers independently detected facial components. On the second level, a geometrical configuration classifier performed the final face detection by combining the results of the component classifiers. Given ....

[Article contains additional citation context not shown here]

B. Heisele, T. Poggio, and M. Pontil. Face detection in still gray images. AI Memo 1687, Center for Biological and Computational Learning, MIT, Cambridge, MA, 2000.

....is a particular regularization approach to regression and classification, and belongs to the class of margin maximizing classifiers. SVMs regularize between empirical loss and the smoothness of the approximating function [20, 5] and have been previously shown to work well for face detection tasks [7]. All SVM classifiers used in this work, unless otherwise noted, were trained with linear kernels and code from [12] 2.2 Prior Work The literature of computer vision is rich with studies of object detection. Among these studies, perhaps the most commonly selected object is the human face. The ....

....faces exclusively. Indeed, successful implementations of component based pedestrian detectors and vehicle detectors are discussed in [13, 10] and [1] respectively. All architectures of component based systems must at some point select which parts to use. Some systems, such as those described in [9, 7], use features which seem naturally salient to humans, such as the eyes, nose, and mouth. Other systems have been designed to learn object parts automatically from the training images [22, 24, 8, 23] The system described in [8] uses 14 features that were chosen automatically using a region ....

[Article contains additional citation context not shown here]

B. Heisele, T. Poggio, and M. Pontil. Face detection in still gray images. A.I. memo 1687, Center for Biological and Computational Learning, MIT, Cambridge, MA, 2000.

....training set contains 6,639 negative examples and 652 positive examples. The testing set contains 1,807 negative examples and 200 positive example. The second data set is a face recognition data set that has been used numerous times at the Center for Biological and Computational Learning at MIT [28, 29], referred to here as faces. The training set contains 2,429 faces and 4,548 non faces. The testing set contains 472 faces and 23,573 non faces. Although RLSC on the entire dataset will produce results essentially equivalent to those of the SVM [21] they will be substantially slower. We ....

B. Heisele, T. Poggio, and M. Pontil, Face detection in still gray images, Technical Report A.I. Memo No. 2001-010, C.B.C.L. Memo No. 197, MIT Center for Biological and Computational Learning (2000).

....The cluttered faces are the synthetic faces set, but with the non face image as background. The real faces are real frontal faces from the CMU PIE face database [13] presenting untrained extreme illumination conditions. The negative test set consists of 4,377 background images consider in [1] to be di#cult non face set. We decided to use a non face set for testing di#erent type from the training non face set because we wanted to test using non faces that could possibly be mistaken for faces. Examples for each set are given in Figure 2 1. Figure 2 1: Typical stimuli used in our ....

....the training set, but they have high ROCs for the test sets. If we use method 2, these features are not picked. Secondly, the training and test non face sets are di#erent types of non faces. The first consists of scenery pictures, while the latter are hard non faces as deemed by an LDA classifier [1]. The features tuned to the training non face set may perform poorly on the test non face set. In Fig. 4 11, we see that the training and test feature ROCs are less correlated for low 0.9 0.95 1 (0.9844) 0.9871) 0.9660) 0.9767) 0.9925) 0.9816) 0.9805) cluttered mixed cl roc roc2 c2 ....

B. Heisele, T. Poggio, and M. Pontil. Face detection in still gray images. CBCL Paper 187/ AI Memo 1687 ( MIT), 2000.

....training set contains 6,639 negative examples and 652 positive examples. The testing set contains 1,807 negative examples and 200 positive example. The second data set is a face recognition data set that has been used numerous times at the Center for Biological and Computational Learning at MIT [28, 29], referred to here as faces. The training set contains 2,429 faces and 4,548 non faces. The testing set contains 472 faces and 23,573 non faces. Although RLSC on the entire dataset will produce results essentially equivalent to those of the SVM [21] they will be substantially slower. We ....

B. Heisele, T. Poggio, and M. Pontil, Face detection in still gray images, Technical Report A.I. Memo No. 2001-010, C.B.C.L. Memo No. 197, MIT Center for Biological and Computational Learning (2000).

....classifier were successfully applied to tasks where the pose of the object was fixed. In [6] Haar wavelet features are used to detect frontal and back views of pedestrians with an SVM classifier. Learning based systems for detecting frontal faces based on a gray value features are described in [14, 13, 10, 2]. Component based techniques promise to provide more invariance since the individual components vary less under pose changes than the whole object. Variations induced by pose changes occur mainly in the locations of the components. A component based method for detecting faces based on the ....

....was calculated relative to the number of non face test images. Because of the resolution required by the component based system, a direct comparison with other published systems on the standard MIT CMU test set [10] was impossible. For an indirect comparison, we used a second degree polynomial SVM [2] which was trained on a large set of 19 19 real face images. This classifier performed amongst the best face detection systems on the MIT CMU test set. The ROC curves in Fig. 3 show that the component based classifier is significantly better than the three global classifiers. Some detection ....

B. Heisele, T. Poggio, and M. Pontil. Face detection in still gray images. A.I. memo 1687, Center for Biological and Computational Learning, MIT, Cambridge, MA, 2000.

No context found.

B. Heisele, T. Poggio, and M. Pontil. Face detection in still gray images. A.I. Memo 1687, Massachusetts Institute of Technology, May 2000.

No context found.

124--129 6. Heisele,B. Poggio,T. Pontil, M. Face Detection in Still Gray Images. In: MIT AI Memo, AIM--1687.

No context found.

B. Heisele, T. Poggio, and M. Pontil, "Face detection in still gray images," A.I. memo AIM1687, Artificial Intelligence Laboratory, MIT, 2000.

No context found.

Heisele, T. Poggio, and Pontil, "Face detection in still gray images," MIT tech report (ai.mit.com), 2000.

No context found.

B. Heisele, Poggio. T., and M. Pontil. Face detection in still gray images. A.I. memo 1687, Center for Biological and Computational Learning, MIT, Cambridge, MA, 2000.

No context found.

Heisele, B.; Poggio, T.; Pontil, M.: Face Detection in Still Gray Images, in MIT AI Memo, AIM--1687, 2000.

No context found.

B. Heisele, T. Poggio, and M. Pontil. Face Detection in Still Gray Images. A.I. Memo 1687, Massachusetts Institute of Technology, May 2000.

No context found.

Heisele, B., Poggio, T., Pontil, M.: Face Detection in Still Gray Images. In: MIT AI Memo, AIM--1687. (2000)

No context found.

B. Heisele, T. Poggio, and M. Pontil. Face detection in still gray images. Technical Report AI Memo 1687 -- CBCL Paper 187, MIT, May 2000.

No context found.

B. Heisel, T. Poggio, and M. Pontil. Face detection in still gray images. Technical Report 1687, Massachusetts Institute of Technology, 2000. AI Memo.

No context found.

B. Heisele, T. Poggio, and M. Pontil. Face detection in still gray images. Technical report, AI Memo 1687, Massachusetts Institute of Technology, 2000.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC