Citations: - 0 self. Abstract Recent work on face identification using continuous density Hidden Markov Models HMMs has shown that stochastic modelling can be used successfully to encode feature information. Keyphrases human face identification stochastic model top-bottom hmm continuous density hidden markov model feature information different parameter frontal image top-bottom scanning stochastic modelling natural order face identification experimental result various hmm parameterisations subjective intuition recent work.
In more recent work [8] the segmentation of the training data was used subjectively to choose the 3. Recognition is carried out via a simple Viterbi recogniser. The recognition process consists of the following steps which are sum- 3. The unknown test image is sampled to generate paper were carried out using the HTK: Hidden Markov an observation sequence Otest. The model with the highest likelihood is selected It is evident that only a subset of the full parame- and this model reveals the identity of the un- ter range needs to be investigated, as inter-parameter known face.
The subjects are either Olivetti test faces are partitioned into rigid, arbitrary regions employees or Cambridge University students. The age with the risk of cutting across potentially discriminat- of the subjects ranges from 18 to 81, with the ma- ing features. In a top-bottom model with no overlap, jority of the subjects being aged between 20 and Subjects sults. Alignment in images of the same subject is pre- were asked to face the camera and no restrictions were served either if the features occupy the exact same po- imposed on expression; only limited side movement sition in all the images or if the features are vertically and limited tilt were tolerated.
Unless the images are preprocessed, the features lighting conditions, but always against a dark back- will normally not be in the same position. Therefore ground. Some subjects are captured with and with- in most cases alignment is preserved only if the ver- out glasses. The images were manually cropped and tical displacement is a multiple of L. Overlap during rescaled to a resolution of 92x, 8-bit grey levels. The the observation sequence will be large.
The histograms of not enough occurrences of model events. In both cases, as 0. If the number of observations in a sequence discussed in section 2. View on IEEE. Save to Library Save. Create Alert Alert. Share This Paper. Background Citations. Methods Citations.
Results Citations. Figures and Topics from this paper. Citation Type. Has PDF. Publication Type. More Filters. Pseudo two-dimensional Hidden Markov Models for face detection in colour images.
This paper introduces the use of Hidden Markov Models HMM as an alternative to techniques classically used for face detection. The result is an efficient projection-based feature extraction and classification scheme for AFR. Soft decisions made based on each of the projections are combined, using probabilistic or evidential approaches to multisource data analysis. For medium-sized databases of human faces, good classification accuracy is achieved using very low-dimensional feature vectors.
Unable to display preview. Download preview PDF. Skip to main content. This service is more advanced with JavaScript available. Advertisement Hide. Discriminant analysis for recognition of human face images. Face-based Authentication First Online: 10 June This is a preview of subscription content, log in to check access.
Baron, R. Bogler, P. Brunelli R. Chellappa, R,. Wilson, C. Craw, I,. Tock, D,. Bennett, A.
0コメント