The Research On Face Recognition Based On Color SIFT And Shape Context Algorithms

In the early1970，s，some scholars began studying face recognition. With theadvancement of computer technology so too have we made significant progress in facialrecognition. Facial image information processing has been widely used in secuirty andinvestigative fields For example, these systems are being used for touirst travel secuirty,public safety and smart phone face recognition.In1999, David G. Lowe proposed the Scale-Invairant Feature Transform algorithm("SIFT" for short). SIFT is used for object recognition, image matching and so on. Thisalgorithm was improved upon in2004，SIFT is used in many aspects, such as objectrecognition, positioning, image stitching, gesture recognition, ifngerpirnting and facialrecognition. SIFT features robustness for scaling，rotation and scale illuminationchange.After analyzing research about face recognition from both domestic and foreignscholars, the new algoirthm, SC-SIFT, will use local feature points common amongfacial recognition systems as the main research direction. The SC-SIFT algorithmimproves upon the descriptors from SIFT by using its unique advantages and applyingit to facial recognition methods. Since SIFT cannot process color images, an additionaldescirptor for color has been added in SC-SDFT and to improve accuracy, SC-SIFT iscombined with the Shape Context algoirthm. Expeirments have shown improvedaccuracy over the original SIFT algorithm. In this thesis, I completed the followingmain work:1. The SIFT algoirthm can only process images in grayscale, however the sampleimages are in color. SIFT was modified to also process color which resulted in greateraccuracy. 2,The improved algoirthm, SC-SIFT, is proposed which combines SIFT and theShape Conetxt algorithm, SIFT and SC descirptors will be combined into newdescirptors and they will be substituted into the SIFT algorithm for matching. Theexpeirment has shown that the new algoirthm has improved accuracy，reaching higherthan90%recognition rate.3.Partial Least Squares was added into the improved algorithm to eliminate falsematching points. Through PLS, SC-SIFT can further improve matching results.