Modified SIFT Descriptors for Face Recognition under Different Emotions

Table 1

Comparative analysis of existing Modified/Improved SIFT techniques.

Method	Description	Recognition rate on JAFFE database			Remarks
		Total images	Recognized images	Recognition rate
Tang et al. [13]; (modified SIFT with EMD)	The algorithm has been implemented to recognize the objects under noise.	213	162	76.05%	Approach is suitable for recognition of objects taken under different view points [13] but does not recognize the facial expressions properly.
Bastanlar et al. [14]; (improved SIFT matching)	The algorithm has been performed on pair of 30 images, viewing four different scenes (indoor and outdoor).	213	157	73.70%	In this method, scale problem of image is eliminated and approach is significant for wide-baseline perspective images pairs and hybrid camera pairs [14].
Ke and Sukthankar [15]; (PCA + SIFT )	The matching performance of the algorithm has been evaluated on Graffiti 6 dataset.	213	161	75.58%	Approach is suitable for yielding concise interesting points, which are invariant over transformations on images caused by changes in camera pose and lightning [15].
Abdul-Jabbar et al. [16] (APCAWT + SIFT)	It was implemented on 200 images of AT&T ORL database and proposed denoised database with accuracy ratio of 85% and 86%, respectively.	213	168	78.87%	It performs well for recognition of face images [16], but it does not yield good result when implemented on JAFFEE database to compare different expressions of face image with only face images of neutral expression (to reduce the database).