Optimizing feature extraction for the camera mouse using genetic algorithms

Authors:
Stephen Karungaru;Minoru Fukumi;Takuya Akashi;Norio Akamatsu
Affiliations:
Institute of Technology and Science, University of Tokushima, Japan;Institute of Technology and Science, University of Tokushima, Japan;Yamaguchi University, Ube, Yamaguchi, Japan;Institute of Technology and Science, University of Tokushima, Japan
Venue:
ICCOMP'06 Proceedings of the 10th WSEAS international conference on Computers
Year:
2006

Citing 5
Cited 0

Detection of regions matching specified chromatic features

Computer Vision and Image Understanding
Structural learning with forgetting

Neural Networks
Unsupervised Texture Segmentation in a Deterministic Annealing Framework

IEEE Transactions on Pattern Analysis and Machine Intelligence
Polynomial-time solutions to image segmentation

Proceedings of the seventh annual ACM-SIAM symposium on Discrete algorithms
Image Segmentation Using Local Variation

CVPR '98 Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition

Quantified Score

Hi-index	0.00

Visualization

Abstract

In this paper, we propose an optimized online facial feature extraction method for a camera mouse using neural networks and genetic algorithms. The method consists of skin color region segmentation and facial features extraction using a neural network whose search is guided by a genetic algorithm. Skin color detection is carried out using a hybrid method based on a YIQ color space threshold and neural network skin color detection methods. The facial features tracked are the eyes and the mouth. A neural network is first trained to detect the eyes and mouth of fixed size and rotation. The genetic algorithm is used to overcome size and orientation invariance by automatic selection of the neural network test samples. Once a feature is detected, it is tracked in the following frames by inheriting the genetic algorithms parameters from the earlier frame. This work was carried out at 7.5 frames per second. Experiments were performed to verify the effectiveness of this method. From the results, an average system accuracy of 95.8% was achieved.