Efficient Retrieval of Images with Irregular Patterns Using Morphological Image Analysis: Applications to Industrial...

Area of Science:

• Computer Vision
• Image Analysis
• Pattern Recognition

Background:

• Image retrieval commonly uses visual content and features.
• Retrieving irregular patterns in industrial/healthcare images is crucial for fault inspection, disease diagnosis, and maintenance prediction.
• Existing methods often rely on deep features, color, shape, or local features.

Purpose of the Study:

• To propose a novel image retrieval framework (ImR) for detecting similar irregular patterns.
• To extract morphological features (DefChars) for enhanced pattern identification.
• To evaluate the framework's performance against various feature extraction methods and distance metrics.

Main Methods:

• The proposed framework extracts a set of morphological features (DefChars) from images.
• Datasets included wind turbine blade defects, COVID-19 CT scans, heatsink defects, and lake ice images.
• Evaluated DefChars against resized raw images, local binary patterns, and scale-invariant feature transforms, using multiple distance metrics.

Main Results:

• The DefChars feature combined with the Manhattan distance metric achieved 80% mean average precision (MAP).
• This combination demonstrated a low standard deviation of ±0.09 across irregular pattern classes.
• The ImR framework outperformed the state-of-the-art deep-learning approach (Super Global) by 8.71% across all datasets.

Conclusions:

• The proposed ImR framework effectively retrieves images with irregular patterns using DefChars and Manhattan distance.
• This method offers superior performance compared to existing feature-metric combinations and deep-learning approaches.
• The framework shows significant potential for applications in industrial inspection and medical diagnostics.