Sparse representation for infrared Dim target detection via a discriminative over-complete dictionary learned online

Area of Science:

• Computer Vision
• Signal Processing
• Machine Learning

Background:

• Traditional dictionaries struggle with ideal sparseness and distinguishing targets from background clutter in natural images.
• Offline learned dictionaries lack adaptability and manual classification hinders efficiency.

Purpose of the Study:

• To propose an infrared dim target detection approach utilizing sparse representation on a discriminative over-complete dictionary.
• To enhance the difference between background clutter and target signals for improved detection accuracy.

Main Methods:

• An adaptive morphological over-complete dictionary is trained online using the K-singular value decomposition (K-SVD) algorithm.
• The dictionary is automatically divided into target and background dictionaries based on sparse decomposition criteria.
• Infrared images are processed using sparse representation and reconstruction residuals to differentiate targets from background.

Main Results:

• The proposed discriminative over-complete dictionary captures significant features of both background clutter and dim targets effectively.
• The method strengthens sparse feature differences between background and target more efficiently than offline methods.
• Experiments demonstrate improved sparsity and enhanced performance in small target detection.

Conclusions:

• The developed online adaptive dictionary approach offers superior performance for infrared dim target detection.
• This method effectively improves the ability to distinguish dim targets from complex backgrounds.
• The approach provides a more robust and efficient solution for sparse representation-based target detection.