In this diploma we focus on recognition of subjects using ear images; more accurately, we focus on evaluation of four algorithms, used as feature extractors, which we can use for either identification or verification. The four algorithms are: Local Binary Patterns – LBP, Local Phase Quantisation – LPQ, Histogram of Oriented Patterns – HOG and Gabor wavelets (also Gabor). Our main tool for evaluating algorithm success is the Receiver Operating Characteristic – ROC curve, but we also use a couple of other mathematical tools.
First, we present similar works, which represent the history of biometric recognition methods. Those first appeared a lot earlier than computers – in the late 19th century, when fingerprints were first used to recognize known criminals.
We proceed with theoretical descriptions of each of the algorithms and comparisons between them. Moving on, we describe our evaluation process, along with the description of the database and the hardware we used.
Furthermore, we show the data, which we used to determine algorithm success and accuracy, showing graphs and processes that we used. We also compare the algorithm success depending on gender and ethnicity of the people in the images.
|
