Avtomatsko iskanje števcev objektov na slikah

Škerjanc, Anže (author), Šajn, Luka (mentor)

Ročno štetje objektov na slikah je časovno zamudno opravilo. Zato si lahko pomagamo z algoritmi, ki avtomatizirajo štetje objektov. Taki algoritmi vsebujejo veliko nastavitev. Nastavitve algoritmov, konstant in rutin zamenjamo s parametri. Vrednost parametrov je treba za optimalno delovanje nad posamezno domeno slik optimizirati. Ročno nastavljanje je težavno, saj parametri med seboj niso neodvisni. Naš cilj je poiskati optimizacijski algoritem, ki bo iz podanega opisa algoritma za štetje in ustrezne domene, v katerem naj bi se ta algoritem uporabljal, poiskal nabor parametrov, ki najbolje štejejo iskane objekte. Zaradi velikega obsega možnih kombinacij (kartezijski produkt domen parametrov) celotni pregled prostora parametrov ni mogoč, zato smo uporabili genetski algoritem. Končni produkt je orodje, ki bo preprosto in uporabno za strokovnjake brez kakršnega koli tehničnega znanja in vedenja o konceptih, ki se skrivajo za temi algoritmi.

• Language: Slovenian
• Keywords: štetje objektov, optimizacija parametrov, genetski algoritem, ImageJ, makro jezik, števec
• Work type: Master's thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2016
• PID: 20.500.12556/RUL-81633
• Publication date in RUL: 21.04.2016

Secondary language

• Language: English
• Title: Automatic search for counters of objects in pictures

Abstract

Manual counting of objects on images is a time-consuming task, therefore we can make us of algorithms that automate object counting. Such algorithms have many settings. We substitute the settings of algorithms, constants and routines with parameters. The value of parameters for individual images has to be optimized for optimal functioning. Manual settings are hard to handle, since the parameters are interconnected. Our goal is to find an optimization algorithm, which will find a set of most suitable parameters for object counting out of the given description of the counting algorithm and the suitable domain, where this algorithm is going to be used. Due to a large number of possible complications (Cartesian product of domains of parameters) a complete overview of the space of parameters in impossible, therefore we used a genetic algorithm. The final product is a tool, which will be simple and useful for experts without any technical knowledge about concepts, hiding behind these algorithms.

• Keywords: object counting, optimization of parameters, genetic algorithm, ImageJ, macro language, counter