The Research Of Eye Detection Algorithm Based On Digital Image Processing

1. SummarizeComputer face recognition is a very active area of research inrecent years. Its applications range is very wide, such as identityauthentication in safety system, video surveillance, target identificationand tracking, as well as facial expression analysis, age analysis, lipreading and so on. Compared with mouth and nose, Eyes are the mostsignificant features of the face, which can provide more reliable, moreimportant message, so eye detection is often necessary to dispose in facerecognition. To do eye recognition, face identification and definitionwill to be done firstly. Face recognition methods generally have thefollowing kinds: First, through the training and detecting with theanalysis method of main weight, namely features face. The methodlocates accurately, but the algorithm is complex and calculation numberis large. Second, Adopted method is to detect people face with themethod of color analysis. But, this method is sensitive to theillumination, and can not apply to gray images. Third, Adopted methodis deformation template category. This method uses elliptical to figurethe contours of the head. Its shortcomings are compute speed is slow,compute time are long. Moreover, the method is to use of neuralnetwork methods of testing people face, its shortcomings are calculateaccount is too large, which can not achieve the weak real-timerequirements. In this paper a new method of detecting eye features infacial images based on gray level projection is put forward. The methodis rapid, simple, can meet the weak real-time applications. In this paper,the studies focus on the identification of a single person which ispositive or semi-angle images.2. The flow of algorithmIn this paper a new method of detecting eye is proposed asfollowed. First, the left and right boundaries of face are localized withvertical gray level projection. Secondly, face region's horizontal graylevel projection is used to localize the top of the head and the middle ofthe nose. Finally, the face structure model is utilized to estimate thepositions of eyes, and the edge features are checked to give the properpositions of eyes. The basic flow algorithm as follows, see figure 1:3. Experiment resultIn order to confirm this article algorithm the validity, to in PICSimage database 100 singles image, as well as 180 singles pictures whichobtains through the scanner, also had many singles image whichabsorbed through video camera to carry on the experiment, Thisalgorithm overall correct recognition rate is 91.2, the small eye and theeye closed situation can also distinguish. The experimental station usespicture size average for 421×539, Programs with the VC++6.0 software,The algorithm on the computer the average running time is 2.09 seconds,Most main time consumes in the edge grouping. When the nervenetwork class algorithm and the distortion template algorithm consumefrequently above a several minutes even hour, This article algorithmsuperiority lies in first through the gradation projection curve definitefacial features area, which causes processing the data quantity to reducegreatly;Then after valid edge examination, the highly effective strategyof edge grouping is used, which guaranteed the right rate and enhancedthe algorithm speed.4. Concluding remarkThis article algorithm localization wrongly appears in the foreheadgradient oversized, as well as the side light too is strong, also has theforehead to occupy the region excessively small situation in the image.Some reasons believed, the algorithm can process under the normalstate the single positive image, also has the very great speed advantage,can basically satisfy the weak real-time application the need. Thealgorithm in computer person face figure recognition, specially inintelligent monitoring aspect, has the very strong practical value.