A comparative study of face landmarking techniques

EURASIP Journal on Image and Video Processing

Table 4 Overview of databases for landmarking studies

	Number	Number	Number of		Control	Reference
	of	of	landmarks ^∗	Modality	tag ^∗∗	works
	subjects	images
Controlled
AR’98 [111]	126	4000	130	color image	e, i, o, f	[31, 123]
						[28, 108]
XM2VTS’99 [112]	295	2360	68	video sequence	o, f, nf, b	[59, 80, 81]
						[41, 47, 85]
						[31, 83, 124]
						[28]
CK + ’10 [114]	123	10,734	68	video sequence	e, f	[28, 44, 108]
Bosp’08 [110]	105	4666	22–24	color, 3D data	e, o, f, nf, r	[28, 90]
Multi-PIE’08 [109]	337	750,000	39–68	color image	e, i, f, nf, r	[47, 105, 123]
						[85, 99, 108]
Uncontrolled
BioID’01 [120]	23	1521	20	gray-scale image	e, i, o, f, nf, s	[81, 82, 105]
						[47, 70, 83]
						[97, 106, 124]
						[28, 31, 108]
Uncontrolled
LFPW’11 [106]	–	3000	29–35	color,	e, i, o, f, nf, s	[97]
				gray-scale
AFLW’11 [121]	–	25,993	21	color	e, i, o, f, nf, r, s	–
AFW’12 [99]	468	205	6	color	e, i, o, f, nf, r, s	–

^*Number of landmarks can vary upon visibility, e.g., 68 landmarks points are not available for profile faces.
^**Acronyms of the control tags are as follows: i: illumination changes, f: facial expression, o: occlusion, r: head rotation, f: frontal face, nf: near-frontal (up to 15° yaw), s: scale change, b: blur. The rightmost column lists representative methods that have used the corresponding database.