BAH, E. Numerické metody pro rekonstrukci chybějící obrazové informace [online]. Brno: Vysoké učení technické v Brně. Fakulta strojního inženýrství. 2019.

Posudek vedoucího

Hnízdil, Milan

The goal of the thesis was to choose, describe, use and compare numerical methods to be able to reconstruct a missing information in data obtained during automotive head lamp de-foging measurements. Student described Mathematical background of used methods for data reconstruction first. These methods were 2D convolution using Gaussian function then interpolation and polynomial regression. Theory of image analysis is described in chapter 3. Fourth chapter deals with image restoration. This chapter is incomplete because it does not describe fully the problematic and solution. There are only compared Gaussian convolution and polynomial regression methods. But any information about partial convolution which was finally used during image restoration is completely missing. Insufficient is also part about image contours map creation. The reader has no chance to understand the problem and its solution. Thesis is written in “good” English with some strange attributes. Sometimes are capital letters in the center of the sentence (especially capital I). References are not sorted correctly. Figures are not referenced in a text and so on. The thesis disappointed me because student created software which is working and will be used in our department also for commercial projects of automotive head lamp de-fogging and de-freezing. Student finished the software and did not have enough time to finish the thesis. Advantage is that student created the code using minimal Matlab functions. All numerical methods were programmed by him-self. Student fulfilled the goal of the thesis.

Posudek oponenta

Procházková, Jana

The diploma thesis presents numerical methods for missing image processing data reconstruction. The work consists of three parts. The first part covers the mathematical background that contains basic definitions. This part is full of mistakes, misspellings, and wrong definitions, e.g. Def. 2.4, 2.5. All theorems are without proofs and do not link to any source literature. In the text, the author mentions that all vectors should be in bold letters, however, in the text, he uses normal letters. In Taylor theorem 2.6, there occurs an unusual member and the other equation (2.19) contains some variables without any explanation. This problem is spread across the work, the work does not contain the type of variables, spaces, range of variables and other common mathematical backgrounds. There is also a misunderstanding of methods; e.g. in Eq. (2.23) with an interpolation: “function f(x) by polynomial h(x) at point (a,b)” is solved by definite integral with a range from a to b. Eq. (2.25) that is also wrong. Some sentences in the text are not finished, they continue with dots “....” A lot of math expressions is wrongly typed. The part 2.10 about regression is really poor, this problem should be in diploma thesis described in much more details. The second part covers the theme Image and image matrix. It is a brief overview of color models, formats and calibration rewritten from English literature. The last part, Image restoration, describes the results of the work. The idea of this work is interesting - how to reconstruct the missing parts of the image of the automotive headlamp. Student wants to test different methods to choose the optimal one. This part has only 8 pages. The student worked in Matlab. In the text is only one kernel matrix h in section 4.4.1. The overview of results is inadequate, tables with testing on one hypothetical matrix P with 5 known values and 4 unknown. The polynomial regression uses a polynomial of degree 7. This is not suitable, there occur problems with a higher degree of the polynomial (Fig. 9). Discussion 4.6 about results takes only 12 lines of text and generalized the result on the behavior of one hypothetical testing. Part 4.7 describes the image restoration process. It consists of one testing case and the results are in two figures. There is no evaluation, no mathematical methods to compute the error, no discussion. Part 4.7.2 is totally unfinished. Fig. 16 is incomprehensible and final table on page 43 without any explanation. In my opinion, this diploma thesis needs a lot of improvement. English level is low, also the spelling and mistakes in sentences’ sense occur really frequently. Literature is also insufficient with problematic notation (different style, missing information). Finally, the list of symbols contains also mistakes - m,n is a matrix, no bold letters for vectors. There is also no explanation of Appendix B in text and file Appendix1.rar in information system is damaged and can not be opened. The theme is sufficient for a diploma thesis but the final work is unsatisfactory. I do not recommend the work for defense and I propose grade F.

eVSKP id 117451