Abstract-This paper presents a multi-step algorithm for noise reduction in hand detection by using partitioning method. The main idea of the partitioning method is to divide an image into equal-size blocks, and obtain the percentage of number of pixels with skin color (skin pixels) in each block. Each of the blocks is labelled as either noise or part of the hand based on the percentage of skin pixels in that block. The image is then refined by repeating the processes to work on smaller blocks in the image with different comparison percentage values. The result of this paper shows partitioning method not only can filter the noise, but it also preserves the shape of the hand better, as compared with the Mathematical Morphology algorithm.Index Terms-Hand detection, noise reduction, human computer interaction, partitioning method. I. INTRODUCTIONImage noise is a major issue yet unavoidable in image processing. Noisy images are produced mainly during image sampling or transmission. Various types of noise such as photo electronic noise, impulse and structured noise present challenging problems in image restoration, particularly in object detection. Several noise reduction techniques have been developed to overcome different types of noise and for different aspects of image restoration.Adaptive noise reduction algorithms such as median filtering, K-nearest neighbor averaging, gradient inverse weighted smoothing, etc. have been introduced in [1]-[3]. Median filtering is particularly effective for reducing impulse noise. It is performed by running through the entire image from one region to another, and replaces each pixel in question with the median of its neighboring entries. However the median based filter often cannot differentiate the thin lines and impulse noises and thus mistakenly removes the thin lines [4].Fuzzy image filtering is a nonlinear processing technique used widely as an alternative to linear adaptive filtering. Mathematical morphology uses fuzzy image filtering to extract image components such as boundaries and skeletons; Manuscript received December 19, 2012; revised February 21, 2013 and also to filter the image noise using erosion processes [5], [6]. However, by using morphology in hand detection, the end effect tends to distort the size of the hand.In our research, we propose to use partitioning method in image noise reduction to improve hand detection for an image with background color close to skin color. Our proposed methodology divides an image equally into blocks. The percentage of number of pixels with skin color (skin pixels) in each block is calculated and compared against a threshold value (skin threshold) to decide if the block is noise or part of the hand. A comparison between partitioning method and morphology algorithm is made, particularly on image noise reduction in hand detection. The result produced by partitioning method displays a cleaner hand image with better shape preservation. II. METHODOLOGYOur project uses partitioning method, which adopts divide-and-conquer strategy to re...

Convex hull vertices extraction from a binary image to detect fingertips always involves multi-step preprocessing algorithm such as edge detection algorithms, before the actual convex hull algorithm can be applied on the image. The pre-processing steps often take up much computational resources. In this paper, we endeavour to reduce the computational resources by introducing a hybrid convex hull algorithm, which is able to extract the convex hull vertices directly from a binary image without going through the edge detection process. Bresenham algorithm is embedded within Jarvis March to replace most of the work required in the edge detection process. In this respect, our pre-processing step is simple and detect only four global extreme points' extraction. The new algorithm yields time complexity of O(N 2 ).

No abstract