Multilevel Segmentation for Food Classification in Dietary Assessment

Area of Science:

• Computer Vision
• Machine Learning
• Image Analysis

Background:

• Accurate object identification and segmentation are crucial for image analysis tasks.
• Existing methods may struggle with perceptual similarity and segmentation accuracy assessment.
• Automated food identification is a key challenge in dietary assessment.

Purpose of the Study:

• To develop a method for automatically identifying and locating perceptually similar objects in images.
• To leverage object class information for assessing image segmentation accuracy.
• To apply the developed method to a dietary assessment tool for food recognition.

Main Methods:

• Combining global and local features to partition segmented objects into perceptually similar classes.
• Generating multiple segmentations per image and learning object classes by merging them.
• Utilizing object class information to evaluate segmentation quality.

Main Results:

• The proposed method effectively identifies and locates perceptually similar objects.
• The approach improves the assessment of image segmentation accuracy.
• Demonstrated utility in a dietary assessment tool for food image analysis.

Conclusions:

• The developed technique offers a robust way to handle perceptual similarity in object recognition.
• This method enhances the reliability of image segmentation by incorporating object class information.
• The application in dietary assessment highlights the practical value of automated food image analysis.