Novel Approach for Constructing Object Contour Conformal to Human Visual Perception

View Statistics Email Alert RSS Feed

Information

Details

Project title

Code/計畫編號

NSC102-2221-E019-056

Translated Name/計畫中文名

建立符合人類視覺感知之物件輪廓的新潁方法

Project Coordinator/計畫主持人

Jung-Hua Wang

Funding Organization/主管機關

National Science and Technology Council

Department/Unit

Department of Electrical Engineering

Website

https://www.grb.gov.tw/search/planDetail?id=3105668

Year

2013

Start date/計畫起

01-08-2013

Expected Completion/計畫迄

31-07-2014

Bugetid/研究經費

771千元

ResearchField/研究領域

資訊科學--軟體

本計畫擬開發之技術有二：(a)影像處理演算法，其特色在於保留符合人類視覺感受之影像邊緣;(b) 建立物件輪廓之演算法，其係基於前述影像處理演算法所獲得之邊緣以建立符合人類視覺感受之影像物件。此二技術期能達到在機器上(如電腦、手機、PDA、醫療儀器等)模擬人類感受影像物件的目的。為達成上揭目的，本計畫擬以兩年時間進行以下工作：(1)以線性與非線性濾波模擬人類視覺感受到之影像資訊;(2)利用資訊理論技術(如亂度分析)克服在傳統邊緣偵測方法上無法對應人類視覺感知邊緣的問題，進而基於區域內複數梯度方向之一致程度以加權的方式 (如貝式事後機率)判斷目標像素是否為感知邊緣上之像素點;(3)基於完形心理學(Gestalt)之連續性(Continuity)與鄰近性(Proximity)原理開發具有學習、概泛能力之回饋式神經網路以建立符合人類視覺的完整物件輪廓。本計畫具有前瞻性，開發完成之技術可將人類感受影像物件的能力(簡化、統整)建立於機器，所建立的物件輪廓可作為後續辨識應用之依據，其應用範圍廣泛，包含雲端影像搜尋、醫療影像、擴增實境、安全監控、車用電子等。This research proposal aims to develop two novel techniques: (a) Perceptual Edges Detection Algorithm (PEDA) by which computing machines (such as computers, mobile phone, PDA, etc.) can mimic the capability of human vision to detect meaningful objects in an input scene, PEDA is characterized in that the extracted edges are conformal to human vision perception; (b) Object Contour Delineating Algorithm (OCDA) by which an object conformal to human visual perception can be accurately established by preserving essential perceptual edges extracted by PEDA. In order to fulfill the foregoing goals, we plan to complete the following works in two years: (1) using linear and nonlinear filters to mimic the capability of human vision system in preserving image objects; (2) overcoming the gradient direction distortion problem encountered in traditional edge detection methods by using the information theory techniques (e.g. Entropy, randomness, etc.). The initial idea to solve this problem is to use a weighted approach (e.g. Bayesian Probability) to determine whether a target pixel is on a perceptual edge or not; (3) based on the continuity and proximity principles in Gestalt psychology, a complete contour of object can likely be constructed, and a recurrent neural network model will be developed for building object contours matched with human vision perception. We anticipate the result of this proposal will find great potentiality in various application fields in the sense that mimicking the perceptual ability of human eyes with computing devices can be applied to fields of medical image, surveillance, automobile electronics, searching image of cloud technology and the like.

Keyword(s)

視覺感知
神經網路
邊緣偵測
物件輪廓
完形心理學
資訊理論
vision perception
neural network
edge detection
object contour
Gestalt
information theory

DSpace CRIS

Novel Approach for Constructing Object Contour Conformal to Human Visual Perception

Details

Description