Human Communication Technologies Lab

Caressing sound and image

A Project by
Sarah Min

Abstract
Downloads
- Papers
Various Interpolation
Various Edge Detection
Current Work/ Other Interesting Stuff
Contact Information

MTC Express is a pressure sensitive multi-point touch pad developed by Tactex Controls Inc.

The touch pad contains 72 sensor points (6 by 12), which are approximately 1cm apart from each other. Due to the low resolution of the device, the reconstructed image is coarse. In order to solve this problem, various methods including Gaussian and Bicubic interpolations are implemented for better extrapolation. Various edge detections are also implemented to capture the boundaries of the object on the touch pad. The project, Caressing Sound and Image, is now being combined with Flow Field by Tim Chen. Please visit the website at http://hct.ece.ubc.ca/research/flowfield for details. On this page, I will only address on the interpolation methods, edge detections, and some interesting representations of the image.

Downloads

Papers

Caressing Sound and Image by Sarah Min

To download the report on the Caressing Sound and Image(2002) prepared by Sarah Min click on the link above.

Various Interpolation

Gaussian Interpolation

The following images are reconstructed with raw input representation method(Nearest Neighbouring Interpolation) and Gaussian Interpolation, respectively. As you can notice, Gaussian Interpolation method rebuilds the image with higher resolutions than Nearest Neighbouring Interpolation Method does. However it still produces blocky artifacts on the image.

A Graphic for the Page

Bicubic Interpolation

Bicubic Interpolation is a well-known method for image transformation. Even though Bicubic Interpolation is more costly in time than Gaussian Interpolation, it visibly produces smoother results.

A Graphic for the Page

Various Edge Detection

Edge Detection is one of the major areas in Image Processing. In this project, it is important to capture the boundary of the object on the touch pad in order to define an impact area in the Flow Field application. To create the best curve fitting points, various edge detection methods including Canny Edge Detector, Threshold Detector, PCA with Kmean Clustering are implemented.

Canny Edge Detection vs Threshold Detection vs PCA

Canny Edge Detector is mainly used for large image files. Due to this characteristic (ie. our image files are in much lower resolutions than normal image files), the first image below, implemented with Canny Edge Detector, is reconstructed with thick lines. Threshold Detector creates a boundary simply by eliminating pixels that are out of limit. The 2nd image below shows the threshold boundary texture-mapped onto a bspline surface that represents the surface of the touchpad. In the last image, Principal Components Analysis(PCA) is applied to a set of clustered date(Kmean Clustering) in order to create an eclipse around the pressed area.

A Graphic for the Page

Current Work/ Other Interesting Stuff

My current interest is to implement real-time virtual clay modelling.

Contact Information

Sarah Min