Page 134 - 2024-Vol20-Issue2

130 | Fathi & Aziz

TABLE I.
COMPARISON BETWEEN SOME OF THE REVIEWED HAND IMAGE DATASETS

Name [10] [11] Refrences [28] [29]
No. of image U-HD 11k hand [27] PolyU Version2 NA
No. of people 15-17/person 1200
11076 IITD Version1 7752 NA
Age 57 190 2300 193 NA
Right- left 18-50 18-75 230 10-55 both
Image size(pixel) both both 12-57 both 1290 * 270
Hand side 1536 * 2048 1600 * 1200 both NA Palme
Capturing device Dorsal - Palme Dorsal - Palme Palme Laptop camera
Purpose Digital camera USB camera 800 * 600 Camera Hand posture
Gender Gender and ID Palme General
Camera
ID

Fig. 2. Framework of the Proposed Work

system that employs deep neural networks and machine learn- to 450*600 pixels. The aforementioned method was utilized
ing techniques to properly identify and track the essential to detect hands and identify significant hand key points that
hand parts in real-time, including the fingertips, palm, and would serve as a guide for the feature extraction procedure.
wrist. A few applications provided by MediaPipe include
hair segmentation, face detection, multi hand tracking, object C. Features Extraction
detection, and monitoring. It refers to 21 joints or knuckle In our proposed work and based on the 21 key points that
coordinates [31] as in Fig. 6 [18]. After the image is captured, were obtained from the MediaPipe algorithm, we extracted
pre-processing is carried out to only get hand area informa- 43 features from the image of right palm with closed fingers
tion. In our study, we utilized the MediaPipe method to rec- (RPC) for each individual in our dataset. As shown in Fig.
ognize and segment the hand region after resizing the photos 7 and Table II, the extracted features were 9(1 to 9) features