Cough Noise Suppressor For Video Conferencing (An Image Processing Approach)

Chung, Soon Thien (2011) Cough Noise Suppressor For Video Conferencing (An Image Processing Approach). Project Report. UTeM, Melaka, Malaysia. (Submitted)

	PDF (24 pages) Cough_Noise_Suppressor_For_Video_Conferencing_(An_Image_Processing_Approach)_-_24_pages.pdf - Submitted Version Download (490kB)
	PDF (Full Text) Cough_Noise_Suppressor_For_Video_Conferencing_(An_Image_Processing_Approach)_-_Full_Text.pdf - Submitted Version Restricted to Registered users only Download (1MB)

Abstract

Video conferencing involved real-time video communication between users present at two different locations. The video, consists of both the image frames and voice, are captured and compressed by a computing unit before the transmission. Coughing produced unpleasant sound while communicating through video conferencing and human being tend to cover their mouth with hand to reduce the volume of the produced unpleasant sound. The possibility of detecting the action of covering the mouth and automatically mute the input audio to suppress the noise produced by coughing is explored in this project. The process of detecting the action of using the hand to cover the mouth, with an image processing approach, can be divided into three steps, namely the human skin pixel classification, skin region extraction and the mouth detection within the extracted skin region. Gaussian Mixture Model (GMM), Bayesian method and Multi Layer Perceptrons (MLP) are three well known methods for skin classification. GMM has been chosen due to the requirement of less memory space compare to Bayesian method and lesser computational cost compared to MLP. Experiment should that reliable human skin classification is needed to produce good result in mouth detection within the classified skin region. Connected component labeling, one of the general region extracting algorithm provided by most of the image processing toolkit, was used to extract the skin region formed by the classified skin pixels. Making use of the extracted region, the region with seven hollows (two eye browns, two eyes, two nostrils, one mouth) is subsequently detected as “face with mouth detected”. The region with only six hollows is detected as “face with mouth not found” and a signal is subsequently send to the system to mute the xii input audio. This allows the coughing noise to be automatically suppressed. The system will send a signal to unmute the input audio when “face with mouth detected” is sensed in the future image frame. Making use of the three identified component, a software prototype has been designed and implemented. Result show that the implemented system is only able to detect both the “face with mouth detected” and “face with mouth covered by hand” in an ideal situation with around 5 frames per second.

Actions (login required)

View Item