Image Target Recognition via Mixed Feature-Based Joint Sparse Representation.

Area of Science:

• Computer Vision
• Machine Learning

Background:

• Image target recognition is crucial for various applications.
• Existing methods often struggle with variations like illumination, deformation, and rotation.
• Data-intensive training requirements limit the applicability of many current approaches.

Purpose of the Study:

• To propose a novel image target recognition approach.
• To enhance robustness against target image variations.
• To develop a data-lightweight classification framework for effective recognition with few training samples.

Main Methods:

• Extracting Gabor wavelet features and deep features using Gabor wavelet transform and Convolutional Neural Networks (CNN).
• Calculating adaptive weights for Gabor and deep features to create mixed features.
• Employing Principal Component Analysis (PCA) for dimensionality reduction of mixed features.
• Constructing a joint feature dictionary using public and private image features.
• Utilizing Sparse Representation based Classifier (SRC) for target recognition.

Main Results:

• The proposed method demonstrates robustness to illumination variation, deformation, and rotation.
• Effective target recognition is achieved even with a limited number of training samples.
• Experimental results indicate superior performance compared to other advanced methods on diverse datasets.

Conclusions:

• The adaptive weighted joint sparse representation approach offers a powerful and efficient solution for image target recognition.
• This method addresses the limitations of data requirements and robustness in existing techniques.
• The framework provides a promising direction for developing more versatile and accurate image recognition systems.