Open Collections

Image reduction and edge-based expansion

University of British Columbia

Image resizing plays an important role in image processing and video stream transmission, specially in the commercial video industries. The most imporstant point of image resizing is that the resized images or videos should keep similar key features such as geometric patterns and the closeness of edges of the original images so that they fit the perception of human vision. Image resizing consists of two parts: image reduction and image expansion. For image reduction purposes, some pixels should be removed or a unified shrinking transforms should be taken. For image expansion, some new pixel values should be added. Image resizing has been an exciting topic in digital image processing due to its extensive uses in industries. Several popular methods for image reduction and expansion methods have been proposed. For image reduction, popular methods are alternative downsampling, average filtering, median filtering and wavelet transform. For image expansion, pixel replication, linear inerterpolation, and cubic interpolation are frequently used. In this thesis, for image reduction, an observation on the image reduction method using Daubechies wavelet transform is made. This observation keeps high frequncy components in the reduced picture and so the reduced picture is obviously sharper than that of the top left corner picture of a wavelet transform. For image expansion, a computationally efficient edge dection method is developed. This new edge detection method generates edge pictures that are very similar to the edge pictures generated by the very known Canny edge detector in closeness, one pixel width and non-zigzagging. Furthermore, the computation of this new edge detection method is much less but more robust than the Canny edge detector. Based on this new edge detection method, a new image expansion method is proposed. This expansion method preserves the edge information very well. The generated picture using our proposed expansion method appears less zigzagged on the edge regions of the picture. The idea is to change the neighborhood pixel values of the edges in a picture expanded by a simple expansion method. The changed pixel values are fitter to human vision than the values generated by the simple expansion method. An important application of our reduction and expansion methods is in video compression. After image reduction, only 25% of the stream source is left for encoding. The encoding process can save 75% of the standard encoding time. Implementation and testing also show that overall 20% to 30% improvements over standard wellknown MPEG2 and H.263 methods are achived in acceptable low bits media stream transmmission using this new method. Above all, our proposed image reduction method gives better quality pictures than those generated by Dauchies wavelet transform, alternative downsampling, average filtering and median filtering methods. Our proposed expansion method outperforms the pixel replication, the linear interpolation and the cubic interpolation methods. It gives crisp and less zigzag pictures. Also these methods used for compression give better compression quality than the standard MPEG2 and H.263 methods.

Thesis/Dissertation

application/pdf

2009-08-14

For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.

http://hdl.handle.net/2429/12235

Electrical and Computer Engineering

University of British Columbia

UBCV

DSpace