A Hybrid Quantum-Classical Algorithm for Resolution Enhancement

Pal, Ankur (2022) A Hybrid Quantum-Classical Algorithm for Resolution Enhancement. Masters thesis, Indian Institute of Science Education and Research Kolkata.

Text (MS dissertation of Ankur Pal (17MS047)) 17MS047_Thesis_file.pdf - Submitted Version Restricted to Repository staff only Download (926kB)

Abstract

When an Optical Character Recognition(OCR) software is given a low-resolution image, it has less information to act on to make a classification, which can produce erroneous results. Hence, we need to find out how to augment said information to help OCR make fewer errors. For this the missing information is needed, which in case of images of text, will come from the character set. If the typeface is known, we take the missing information from each character of the character set, one by one, and note which character provides the missing information best. This can be measured by taking the confidence value for that character by the OCR. The process of providing the missing information to the low-resolution image was done by a quantum algorithm called the Fixed-Point Quantum Search by Lov K. Grover as it was found to be the most suitable. Hence, it was applied in two ways: a less efficient 1-qubit operation, which produces results which are closer to high-resolution image, and the more efficient 4-qubit operation. The 4-qubit operation can be further generalised as the n²-qubit operation, where n² is the scale-up factor. The confidence value was obtained using MATLAB, however results were unsatisfactory. Upon analysis, the data suggested a bias in the OCR, hence a simple metric called “Match Percent” was defined. By this metric, all of the characters were correctly identified, serving as a proof of concept. Upon complexity analysis, it was found that, for the general case where one pixel of grayscale image is represented by 8 bits, it was computationally much expensive to perform n2-qubit operation on a classical computer, even when taking multiple measurements needed in quantum algorithm were taken into account. Although it is not currently possible to implement it, the quantum algorithm will be even more comparatively efficient if applied for the total image at once. Finally, the ways in which we can improve this algorithm and its limitations are discussed.

Actions (login required)

View Item