The Acquisition-Assumption-Assessment-Adaptation-Accumulation Paradigm for Mining Multiple Structures and Detecting Multiple Objects
(L. Xu)

Tasks of detecting multiple objects on an image or mining multiple dependence structures from data are widely demanded in computer vision, image recognition, data mining as well as various biometric based inspections. These tasks consists of two types of problems. The first is to segment each object or pattern structure not only from others and but also from noise and background. The second is to recognize each object and to model each structure. However, the performance of solving one type depends on closely the performance of solving the other. This coupling makes these tasks difficult to tackle. Finite sample acquisition, multiple candidate assumption, criterion guided assessment, local search adaptation, and global consensus accumulation are five classical problem solving strategies that have been often used in human intelligent activities. This project aims at adopting these strategies to automatic problem-solving via an integrated Acquisition-Assumption-Assessment-Adaptation-Accumulation paradigm for detecting multiple objects on an image and mining multiple structures from data, with effective algorithms developed for real shape detection tasks on images, especially on biomedical images.