Vector Quantization

Demo for Artificial Neural Network Course

December, 2015

Vector Quantization on RGBD image using Simulated Grossberg Network and Modified ART2 Network

Using scientific computing method to simulate Grossberg Network and modified Adaptive Resonance Theory (ART2), for the purpose of Vector Quantization, i.e., image compression of RGBD images captured by depth-sensing camera, e.g., Kinect.

Vector Quantization (VQ) is known to be a useful technique in lossy data compression. The Grossberg network is a self-organizing continuous-time competitive network, which can be used for data normalization, contrast enhancement and noise suppression. ART2, is a cognitive and neural theory to attend, categorize, recognize, and predict data. The combination of these two neural nets can be used to process the image data for the purpose of vector quantization. Recently, RGBD textures are increasingly found due to the development of depth-sensing cameras, it is vital to apply image processing methodology upon RGBD images.

Report PDF: