Computer implemented method for sign language characterization
View Patent ↗A sign language recognizer is configured to detect interest points in an extracted sign language feature, wherein the interest points are localized in space and time in each image acquired from a plurality of frames of a sign language video; apply a filter to determine one or more extrema of a central region of the interest points; associate features with each interest point using a neighboring pixel function; cluster a group of extracted sign language features from the images based on a similarity between the extracted sign language features; represent each image by a histogram of visual words corresponding to the respective image to generate a code book; train a classifier to classify each extracted sign language feature using the code book; detect a posture in each frame of the sign language video using the trained classifier; and construct a sign gesture based on the detected postures.
1. A computer-implemented method of recognizing sign language, the method comprising:
detecting, via circuitry, one or more interest points in an extracted sign language feature, wherein the one or more interest points are localized in space and time in each of a plurality of images acquired from a plurality of frames of a sign language video including the extracted sign language feature, wherein the detecting is carried out using a Scale Invariant Features Transform (SIFT) descriptor and the interest points represent corners in each image;
applying a digital filter to determine one or more extrema of a central region of the one or more interest points;
associating one or more features with each interest point of the one or more interest points using a neighboring pixel function;
clustering, via the circuitry, a group of extracted sign language features from the plurality of images based on a similarity between the extracted sign language features according to the associating;
representing each image of the plurality of images by a histogram of visual words corresponding to the respective image to generate a code book;
training, via the circuitry, a classifier based on labels assigned to the plurality of images to classify each extracted sign language feature using the code book;
detecting, via the circuitry, a posture in each frame of the plurality of frames of the sign language video using the trained classifier;
constructing, via the circuitry, a sign gesture based on the detected postures, and
identifying text words that correspond with the sign gesture and presenting the text on a display so as to ease communication between deaf people and non-deaf people.
2. The method of claim 1 , wherein the digital filter includes a Difference of Gaussian (DoG) filter to detect a blob-like shape in each image as the one or more interest points.
3. The method of claim 1 , wherein the digital filter includes a Harris Laplace.
4. The method of claim 1 , further comprising:
training a Support Vector Machine to classify each frame of the sign language video.
5. The method of claim 1 , further comprising:
classifying each sign gesture using a K-Nearest Neighbors (K-NN) classifier.
6. The method of claim 1 , further comprising:
splitting a sequence of postures, and generating a split histogram of the postures for the respective split sequence of postures; and
concatenating the split sequence of postures to preserve an order of the sequence of postures in the respective sign gesture.
7. The method of claim 6 , further comprising:
normalizing the split histogram of the sequence of postures of the sign language video to account for a difference in signing speed.