Real Time DSP based Identification System using Content-Based Imaging Techniques

In this thesis work we propose a technique for the design and development of an automatic visual identification system using content-based matching.The proposed system is implemented using an interconnection of four sub-system :

1. sensing,
2. data acquisition,
3. feature(content)extraction and
4. feature analysis.

This system is based on indentification of images using content based matching of a query image with those of the datdbase images. Quary images is the on-line grabbed image by the area scan CCD cameras.Database (off-line) is prepared for all expected query images by extracting relevent features.

Image histogram(gray level) is used for feature extraction.The computational complexity & storage requirements are reduced by decomposing the histogram using Wavelet Tranform.First and second moments of these wavelet coefficents is used as features.The root mean square(rms) metric ic is used to compute the distance between the query image with that of the datdbase images.

Although the system is designed to inspect steel slab for surface defects but laboratory evaluation of this system gives excellent performance with generl textured and non-textured images also. A setup of high speed Digital Signal Proc essors is used to keep up with the required real time throughput rates of 1024K pixels/sec(1 m wide steel slab moving at the rate of 1 m/sec).

[Thesis List] [ME Home Page]