Related Experiment Videos
Design-based texture feature fusion using Gabor filters and co-occurrence probabilities
1Department of Systems Design Engineering, University of Waterloo, Waterloo, ON N2L 3G1 Canada. dclausi@engmail.uwaterloo.ca
Summary
This study introduces a new texture recognition method by combining Gabor filter and grey level co-occurrence probability (GLCP) features. The fused approach enhances image segmentation accuracy, even with noisy images.
Area of Science:
- Computer Vision
- Image Processing
- Pattern Recognition
Background:
- Texture recognition is crucial for image analysis.
- Existing methods using individual Gabor filters or GLCP have limitations in capturing comprehensive texture information.
- Noise in images often degrades the performance of texture recognition algorithms.
Purpose of the Study:
- To develop a novel fused feature set combining Gabor filter and GLCP features for enhanced texture recognition.
- To evaluate the effectiveness of the fused feature set in improving image segmentation accuracy.
- To assess the robustness of the fused feature set against image noise and the curse of dimensionality.
Main Methods:
- A design-based method was developed to fuse Gabor filter features (capturing low-mid frequencies) and GLCP features (capturing high frequencies).
- Feature space separability and image segmentation classification rates were used as evaluation metrics.
- The performance of the fused feature set was compared against individual feature sets, including under various noise conditions.
Main Results:
- The fused feature set demonstrated superior feature space separability and higher segmentation accuracies compared to individual feature sets.
- The fused features showed improved performance on noisy images across different noise magnitudes.
- The 48-dimensional fused feature set was found to be resistant to the curse of dimensionality.
- Principal Component Analysis (PCA) was acceptable for feature reduction, while feature contrast method significantly reduced accuracy.
Conclusions:
- The proposed fused feature set effectively combines complementary texture information from Gabor filters and GLCP.
- This fusion method offers a robust and accurate approach for texture segmentation, outperforming individual feature extraction techniques.
- The method provides a viable solution for improving texture recognition performance, particularly in the presence of noise.