[Retracted] Cloud Data Access Prevention Method in Face Recognition Technology Based on Computer Vision

Table 3

Gradient distribution comparison.

	Average gradient of each subregion (%)				Distribution of the top 25% maximum
CNN	0.35	3.8	6	11.2	0	0.7	0.24	27.6
	0.9	5.74	8.4	21	0	0.01	0.1	60.4
	1.8	11.7	5.7	13.6	0	0.04	0	11
	0.37	3.6	1.9	3.9	0	0	0	0
Add occlusion trained CNN	0.37	5.4	5.5	2.6	0	2	1.5	0.6
	1.3	10.5	20.34	17.8	0	2.4	32.6	55.5
	1.07	6.3	16.5	10.7	0	0	2.74	2.7
	0.05	0.5	0.7	0.54	0	0	0	0
MRCNN	1.5	5.2	5.8	1.12	0.8	4	2.9	0.5
	4.8	16.4	15.5	3	5.1	14.8	15.3	2.1
	3.9	15.1	14	4	4.1	14.1	19.5	10.4
	1.5	3.9	3.1	1.1	0.45	2	3.9	1.2