Occlusion detection prior to face recognition using structural feature extraction
| Autor: | Rohit Tayade |
|---|---|
| Rok vydání: | 2017 |
| Předmět: |
Pixel
Computer science business.industry Feature extraction ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION 020206 networking & telecommunications Pattern recognition 02 engineering and technology Occlusion detection Facial recognition system Regression Support vector machine ComputingMethodologies_PATTERNRECOGNITION Face (geometry) Linear regression 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing Artificial intelligence business |
| Zdroj: | 2017 International Conference on Intelligent Computing and Control Systems (ICICCS). |
| DOI: | 10.1109/iccons.2017.8250722 |
| Popis: | There has been enormous amount of research on face recognition which consists of modification of state of art face recognition methods to make it able to work under partial occlusion, illumination/pose changes. Those modifications came up with additional computations which have compromised with execution time for accuracy. The modified version of LRC(linear regression classification) known as CRC (census regression based classification) grabbed an attention of researchers through its recognition performance. These CRC generates census transformed test and training images then estimates importance of each pixels which takes twice amount of time than execution time of LRC. The goal of this research is to improve execution performance of CRC(census regression classification) by eventually switch to LRC face recognition system for non-occluded face by detecting whether the face is occluded or not prior to face recognition using structural features of face and SVM. Experimental analysis shows that proposed occlusion detection method has 96% accuracy for artificially generated occlusions of face images of ORL face database and execution time has been reduced by more than 50% for non-occluded faces. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|