Editing DCT-Based Texture Classification Using Soft Computing Approaches

Title: DCT-Based Texture Classification Using Soft Computing Approaches

Research Question: Can soft computing methods, such as artificial neural networks and neuro-fuzzy systems, effectively classify textures using Discrete Cosine Transform (DCT) coefficients?

Methodology:
1. Preprocessing: Convert color images to grayscale.
2. DCT Transformation: Apply DCT to grayscale images to obtain coefficients.
3. Feature Extraction: Use DCT coefficients as features for texture classification.
4. Soft Computing Models: Train artificial neural network (ANN) and neuro-fuzzy system (NFS) using backpropagation and evolving fuzzy neural network algorithms, respectively.
5. Classification: Use 80% of texture data for training, and the remaining 20% for testing and validation.
6. Performance Comparison: Compare the performance of ANN and NFS in terms of classification accuracy.
7. Training Epoch Analysis: Analyze the effect of prolonged training on the performance of ANN.

Results:
1. The proposed neuro-fuzzy system (NFS) performed better than the artificial neural network (ANN) for texture classification.
2. Prolonged training of the ANN improved its performance, but the NFS maintained its superiority.

Implications:
1. The use of DCT coefficients as features for texture classification can reduce the computational complexity and storage requirements.
2. Soft computing approaches, such as NFS and ANN, can effectively classify textures using DCT coefficients.
3. The NFS is a more efficient classifier for texture classification compared to ANN, especially when dealing with complex textures.
4. The analysis of training epochs highlights the need for careful selection of training parameters to balance accuracy and computational cost.

Link to Article: https://arxiv.org/abs/0405013v1
Authors: 
arXiv ID: 0405013v1

[[Category:Computer Science]]
[[Category:Dct]]
[[Category:Classification]]
[[Category:Ann]]
[[Category:Texture]]
[[Category:Nfs]]