BATURE, J. Obrazové deskriptory a jejich použití pro detekci objektů v obrazech [online]. Brno: Vysoké učení technické v Brně. Fakulta strojního inženýrství. 2022.

Posudek vedoucího

Procházková, Jana

I worked with Mr. Bature from September 2021. His work was image descriptors and their usage for person detection. He had short time to finish this topic because he started completely new topic. We had appointments every 2 weeks, he worked regularly according to my guidance. His work consisted of completing three tasks. The first part was devoted to a summary of the literature on different types of detectors and descriptors. The graduate student, Mr. Bature, selected interesting types of image descriptors (based on colour or histogram and edges) and modern detectors such as the SURF or Harris detector. He gave a simple description and comparison of their use. The thesis contains 70 citations to scientific literature sources. The second task concerned a specific Viola-Jones detector, which is the basis for simple training and detection of facial parts. It requires a so-called integral image, which the student described. He used only integrated function of Matlab software to implement it. This part should be more precise and should contain more mathematical background with better notation. The last task was the selection of one of the detectors and the actual face detection program. The student chose a method based on eigenvectors and the his diploma thesis contains a theoretical description of the algorithm and also the implementation in Matlab. I have a few comments on the development of the thesis. Some parts are heavily taken from the literature (with proper citation of course), but this makes the whole text difficult to read. The student had problems with the mathematical notation of the individual algorithms, and there are sometimes inconsistencies. Part 4.3 is the mathematical description of proposed algorithm. The notation is not very clear with mistakes, some explanation is missing, there is problem with size of matrix multiplication on page 33-34. The implementation part of 5 is written in colloquial language and could use more formalism. Also, the student unfortunately did not manage to add time consuming tests on multiple image databases with hard e.g. lighting conditions and their evaluation. But the program works properly on testing set. Considering that the thesis was written in 6 months, the student reached a level that is sufficient for the defence and therefore I approve the thesis and I propose final mark: E

Posudek oponenta

Druckmüllerová, Hana

The master thesis deals with image descriptors for face detection and recognition. It first describes various image descriptors and feature detectors and descriptors, then the Viola-Jones algorithm and finally the methods based on eigenvectors. The descriptions are rather wide than deep and what I miss the most is consistent denotation and more mathematical formulations with formulae, especially in sections 2 and 3. At many places it is clear that the author does not know well enough the mathematical language and conventions (see e.g. formula 3.2 on page 23, the first formula on page 27 and generally pages 31-33). I appreciate that the author studied many sources to give the overview of the topic and also the graphics and typesetting is of rather good quality. Since an amount of work that was done I suggest the evaluation by grade E if the questions are answered satisfactorily.

eVSKP id 139584