PARSEG: a computationally efficient approach for statistical validation of botanical seeds' images

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Binary segmentation validation is computationally intensive due to large image pixel counts.
• Current methods like human recognition and automated validation are time-consuming and resource-heavy.

Purpose of the Study:

• To introduce PARSEG, a novel statistical method for validating binary segmentation outputs.
• To reduce the computational complexity and time required for image validation.

Main Methods:

• PARSEG employs a four-step procedure: Partitioning, Random Selection, Estimation, and Generalization.
• It utilizes binary classifiers and an objective function to select an optimal pixel subset for validation.
• The method performs statistical validation on selected pixels to represent the entire image.

Main Results:

• PARSEG significantly reduces the number of pixels needed for validation (e.g., 4% for 13 million pixel images).
• It achieves validation precision comparable to using the entire image.
• Computational time for validation is reduced by approximately 90%.

Conclusions:

• PARSEG offers an effective and computationally efficient approach to binary image validation.
• The method is particularly beneficial for large-scale image datasets in fields like seed recognition.
• It demonstrates a substantial improvement in processing time without sacrificing validation accuracy.