Image Processing Techniques (implemented in MatLab) for ECE 181B : Computer Vision (Winter 2010)
A. Face Recognition with Eigen Faces
B. Object Detection with Bag of Features
C. Homogeneous Transformation
D. Scale Invariant Feature Transform
E. Corner Detection (Harris Corner Detector)
Audio Visual Piece for MAT 200B : Music and Technology (Winter 2010)
This composition was an exploration in mapping of one digital domain into another. The basic process involved creating visuals in Flash, importing them in MatLab and then converting each frame into a corresponding Sound based on its frequency constituents. The conversion was done using Fast Fourier Transform. It involved a lot of experimentation, some expected results and some unexpected. The final result is an audio-visual whole that is inherently in sync with both components. Initial idea for this project stemmed from experiments with inverse conversion of Sound sonograms.
Showcased at the MAT EoYS Concert- May, 2010.
Interactive Multimedia Project (MAT 594O : Sensors)
This project was an exploration to link ideas in Human Computer Interaction and Computational Geometry to create an engaging audio-visual environment. The concept of the project was to trigger a series of 3D forms (based on the SuperFormula proposed by Johan Gielis) transformations & Sound (Vibrato) transformations in Virtual (3d) Space. All action is triggered through a series of hand gestures (Arduino w/ Accelerometer), motion of the hand rotation measured (X,Y,Z) axis and the up/down quick movements to change the SuperShape. Further adjustments to signal received in the system can be used to create more expressive transformations of the SuperShapes.
The user/viewer will be presented an interface (Arduino attached to either right/left hand) that will engage the system, stimuli of site, sound and motion. The areas of HCI, Computer Graphics and Electronic Music were concurrent areas for research.
Check out the video:
SIFT or Scale Invariant Feature Transform is a nifty transform for object recognition using images , to make them scale invariant and to some extent illumination invariant as well. The first image shows a lamp being recognized in a larger photoof the room.
The second image (of the cameraman) shows how even after an image is scaled, rotated and tampered with its brightness a bit, the algorithm still matches it with the original one. This was implemented in MatLab for a Computer Vision class.
We (group of 4 people) had a lot of fun experimenting at every stage to check what the results were. Since the algorithm works on an image on various Octaves and different scales, at one stage we had more than a dozen images to look at. The last image is a screenshot of that one time!