Multilevel image segmentation with application in dietary assessment and evaluation

Fengqing Zhu, Purdue University

Abstract

This thesis describes methods for image analysis, including image calibration, image segmentation, features extraction and classification with emphasis on the segmentation of non-rigid objects. We developed a color fiducial marker to correct colors of unknown image illumination that appear in the scene so that this information can be used by image analysis tasks in our dietary assessment system. We proposed and implemented a multiple hypothesis segmentation technique to select optimal segmentations based on confidence scores assigned to each segment and showed improvements in both segmentation and classification accuracy. We demonstrated the use of active contour models to refine image segmentation. We examined both quantitative performance and classification based evaluation to validate proposed segmentation methods. The proposed image analysis methods were developed for a dietary assessment application. There is a growing concern with respect to chronic diseases and other health problems related to diet including obesity and cancer. The need to accurately measure diet (what foods a person consumes) becomes imperative. Dietary intake provides valuable insights for mounting intervention programs for prevention of chronic diseases. Measuring accurate dietary intake is considered to be an open research problem in the nutrition and health fields. We describe a novel mobile telephone food record that can provide an accurate account of daily food and nutrient intake by analyzing images of the food eaten by a user. Our approach includes the use of image analysis tools for identification and quantification of food that is consumed at a meal.

Degree

Ph.D.

Advisors

Delp, Purdue University.

Subject Area

Computer Engineering|Engineering|Electrical engineering|Medical imaging

COinS