Lipreading method using color extraction method and eigenspace technique

Abstract

SUMMARYThis paper describes a lipreading system using a color extraction method and several eigenspace techniques. To detect a precise mouth position, the system first finds the rough position of the mouth by the color extraction method and then determines the precise position using an eigentemplate technique. To describe the mouth shape change, we developed an eigen waveform technique. The detected precise mouth position improves the accuracy of the eigen waveform, which results in a high recognition rate. We constructed an automatic lipreading system and an experiment revealed that this system recognizes 85% of 35 words and 69% of 100 words.

4D Automatic Lip-Reading for Speaker's Face Identification

“…They weren't using spatial homogeneity, pixel-based algorithms attempted to be faster than region-based approaches from time to time generating coarse segmentation results and no re-correctly classify for pixels is noisy .The pixel appearance is relative to the colour as define and the local neighborhoods of pixels was included. Lip pixels segmentation were practiced by a Bayesian classifier ,this introduced by [56] which uses Gaussian Mixture Models (GMM) that are estimated by using the Expectation Maximization algorithm (EM) , refer to [55] and their normalized the R and G components of the RGB space by using the intensity of the pixel. The normalization was an illumination invariant to RGB.…”

Computer Science & Information Technology

“…The pixel appearance is relative to the colour as define and the local neighborhoods of pixels was included. Lip pixels segmentation were practiced by a Bayesian classifier ,this introduced by [56] which uses Gaussian Mixture Models (GMM) that are estimated by using the Expectation Maximization algorithm (EM) , refer to [55] and their normalized the R and G components of the RGB space by using the intensity of the pixel. The normalization was an illumination invariant to RGB.…”

4D Automatic Lip-Reading for Speaker's Face Identification

“…The pixel appearance is relative to the colour as define and the local neighborhoods of pixels was included. Lip pixels segmentation were practiced by a Bayesian classifier ,this introduced by [56] which uses Gaussian Mixture Models (GMM) that are estimated by using the Expectation Maximization algorithm (EM) , refer to [55] and their normalized the R and G components of the RGB space by using the intensity of the pixel. The normalization was an illumination invariant to RGB.…”