Automatic retrieval of shoeprint images using blocked sparse representation

Area of Science:

• Forensic Science
• Computer Vision
• Pattern Recognition

Background:

• Shoe marks are crucial evidence in criminal investigations.
• Accurate matching of shoeprints aids in criminal identification.
• Existing methods for shoeprint retrieval have limitations in accuracy and robustness.

Purpose of the Study:

• To develop a robust and efficient method for matching shoeprints.
• To improve the accuracy of criminal identification using shoe mark evidence.
• To address challenges in retrieving partial and distorted shoeprints.

Main Methods:

• A blocking sparse representation technique was developed.
• Queried images were divided into blocks for feature extraction.
• Sparse representations were extracted using an approximate ℓ1 minimizing method.
• A referenced database was categorized into two parts with separate dictionaries.
• Total errors were measured by resetting coefficients of other classes to zero.

Main Results:

• The proposed method demonstrated high effectiveness and efficiency in retrieving whole, partial toe, and heel shoeprints.
• Accurate identification rates reached 99.47% for whole shoeprints, 80.53% for partial toe, and 79.47% for partial heel prints in the first rank.
• The method showed superior performance compared to Wright's sparse representation, Fourier transform, Gabor transform, and SIFT descriptors.
• The proposed technique proved resistant to rotation and scale distortions.

Conclusions:

• The blocking sparse representation technique is a highly effective and efficient method for shoeprint retrieval and identification.
• This method significantly enhances the accuracy of forensic investigations by improving shoeprint matching capabilities.
• The technique's robustness against distortions makes it a valuable tool for real-world crime scene analysis.