Development of background subtraction algorithm for biometric identification
This thesis presents an improved approach for an automatic face detection system. Segmentation of novel or dynamic objects in a scene can be achieved using background subtraction or foreground segmentation. This is a critical early step in most computer vision applications in domains such as surveil...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Subjects: | |
| Online Access: | http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/63458/1/Page%201-24.pdf http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/63458/2/Full%20text.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This thesis presents an improved approach for an automatic face detection system. Segmentation of novel or dynamic objects in a scene can be achieved using background subtraction or foreground segmentation. This is a critical early step in most computer vision applications in domains such as surveillance and human-computer interaction.
The proposed system consists of three parts. In the first part, the use of
background subtraction algorithm to deal with the problem of lighting changes,
shadows and repetitive motions. All previous implementations fail to handle
properly one or more common phenomena, such as global illumination changes,
shadows, inter-reflections, similarity of foreground color to background and
non-static backgrounds (e.g. active video displays or trees waving in the wind).
The proposed method is a background model that uses per-pixel, time-adaptive
and Gaussian mixtures in the combined input space of pixel neighborhood and
luminance invariant color. This combination in itself is novel. In the second part,
another technique known as morphological erosion and dilation operators are
used to remove the noise in the resulting binary image to improve the accuracy.
The third part is accomplished by using a new technique to locate the face position
in the image and extract ilfor recognition and identification purposes.
The algorithm has been tested in several different lighting conditions and
environments. The experimental results show that the method possesses much
greater robustness to problematic phenomena than the prior state of the art
methods, without sacrificing real-time performance, making it well-suited for a
wide range of practical applications in video events which requiring detection in
real-time.
The experimental results in real time applications show the robustness, reliability
and efficiency in fhe proposed approach; they can accurately detect and extract
human face 98% of the time, with the ability to detect the face of different types of
people gender, skin color and head attire. The proposed algorithm can be executed
at 30 to 35 FPS for an image size of 320 x 240 pixel, which is much better when
compared with any other real time applications. |
|---|