Font Size: a A A

Image reconstruction, recognition, using image processing, pattern recognition and the Hough transform

Posted on:1993-12-11Degree:Ph.DType:Dissertation
University:The Ohio State UniversityCandidate:Seshadri, M. DFull Text:PDF
GTID:1478390014497437Subject:Physics
Abstract/Summary:
In this dissertation research, we have demonstrated the need for integration of various imaging methodologies, such as image reconstruction from projections, image processing, pattern and feature recognition using chain codes and the Hough transform. Further an integration of these image processing techniques have been brought about for medical imaging systems. An example of this is, classification and identification of brain scans, into normal, haemorrhaged, and lacunar infarcted brain scans.;Low level processing was performed using LOG and a variation of LOG. Intermediate level processing used contour completion and chain encoding. Hough transform was used to detect any analytic shapes in the edge images. All these information were used by the data abstraction routine which also extracted information from the user, in the form of a general query. These were input into a backpropagation, which is a very popular supervised neural network. During learning process an output vector was supplied by the expert to the neural network. While performing the neural network compared the input and with the help of the weight matrix computed the output. This output was compared with the expert's opinion and a percentage deviation was calculated. In the case of brain scans this value was about 95%, when the test input vector did not vary, by more than two pixels with the training or learning input vector.;A good classification of the brain scans were performed using the integrated imaging system. Identification of various organs in the abdominal region was also successful, within 90% recognition rate, depending on the noise in the image.
Keywords/Search Tags:Image, Recognition, Using, Brain scans, Hough
Related items