DeepLumina: A Method Based on Deep Features and Luminance Information for Color Texture Classification

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Color texture classification is crucial for object recognition in natural scenes.
• Deep learning architectures show promise for complex texture pattern recognition.
• Integrating color and texture information is essential for accurate image analysis.

Purpose of the Study:

• To propose DeepLumina, a novel method for efficient color texture classification.
• To leverage deep convolutional neural network features and luminance information.
• To explore the effectiveness of the RGB-luminance color domain for texture analysis.

Main Methods:

• Utilized ResNet101 pretrained models for capturing convolutional neural network features.
• Incorporated luminance information from the Y channel of the YIQ color model.
• Employed a support vector machine (SVM) for the final classification task.
• Evaluated the method in the RGB-luminance color domain.

Main Results:

• Achieved 90.15% accuracy on the Flickr Material Dataset (FMD).
• Obtained 73.63% accuracy on the Describable Textures dataset (DTD).
• Demonstrated computational simplicity and reduced computation time compared to other methods.

Conclusions:

• DeepLumina offers a promising and efficient approach to color texture classification.
• The integration of luminance information alongside RGB features significantly improves performance.
• The method provides a computationally efficient solution for real-world texture analysis tasks.