目标协同分割与识别技术的研究

项目名称： 目标协同分割与识别技术的研究

项目编号： No.61471321

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 无线电电子学、电信技术

项目作者： 于慧敏

作者单位： 浙江大学

项目金额： 80万元

中文摘要： 人类能够在复杂场景中精确分割和识别目标，但这对于机器视觉却是一种严峻的挑战。针对这一难题，本项目提出了图像协同分割与识别技术的研究。本项目试图从内在工作机理和更微观层面上模拟自底向上与自顶向下这两个过程之间的协调工作机理，来研究图像的协同分割与识别技术，提高其处理能力。希望通过本项目的研究能够进一步理解自底向上与自顶向下这两个过程之间的协调工作原理，减少这两个过程之间鸿沟，并推动其在计算机视觉中的应用。本项目研究属于计算机视觉、机器学习、认知科学和数学等多学科交叉，项目将围绕如何理解和模仿自底向上和自顶向下这两个过程的协同工作机理、先验数据统一表达问题、多层特征推理与重建、基于推理和重建的图像协同分割、深度学习等关键问题与技术,来研究图像协同分割与识别计算模型及其相关问题与技术。本项目将实现一个图像识别系统的DEMO，用于互联网或物联网中特定或不良图案图像的识别。

中文关键词： 目标分割；目标识别；深度学习；先验知识

英文摘要： Human can segment and recognize objects precisely in complex scenes. However, it has proven to be a severe challenge for machine vision systems. Aimed at this challenge, this research pursues a cooperative solution to the problems of object segmentation and recognition. This study attempts to provide a different solution to cooperative object segmentation and recognition by simulating the mechanism of the coordination between bottom-up and top-down on the inner mechanism and a more micro level.We hope to further understood the mechanism of the coordination between bottom-up and top-down through this study, and want to be able to reduce the gap between these two processes and to promote its application in computer vision. This project belongs to computer vision, machine learning, cognitive science and mathematics of interdisciplinary，and will be to study the computing model of cooperative object segmentation and recognition and related issues surrounding some key issues such as how to understand and mimic the mechanism of the coordination between bottom-up and top-down, prior data uniform representations, multi-layer features based inference and reconstruction，inference and reconstruction based cooperative object segmentation and deep learning. This project will implement an image recognition system DEMO, for the particular or bad information image recognition in the Internet or Internet of things.

英文关键词： Object segmentation;Object recognition;Deep learning;Prior knowledge