<?xml version="1.0" encoding="utf-8"?>
<export-example>
  <doc>
    <id>298</id>
    <completedYear>2021</completedYear>
    <publishedYear/>
    <thesisYearAccepted/>
    <language>eng</language>
    <pageFirst/>
    <pageLast/>
    <pageNumber>50</pageNumber>
    <edition/>
    <issue/>
    <volume/>
    <type>bachelorthesis</type>
    <publisherName/>
    <publisherPlace/>
    <creatingCorporation/>
    <contributingCorporation/>
    <belongsToBibliography>0</belongsToBibliography>
    <completedDate>--</completedDate>
    <publishedDate>--</publishedDate>
    <thesisDateAccepted>--</thesisDateAccepted>
    <title language="eng">Comparing and controlling image colorization methods for realistic reconstructions</title>
    <abstract language="eng">Machine Learning based grayscale image colorization methods have, in recent years, become good enough such that only few user selected color points are necessary for realistic colorizations. Converting images to grayscale and keeping only few color cues, while further compressing the grayscale image, could be a viable way to reduce file sizes for bandwidth constrained scenarios even further.&#13;
&#13;
In this thesis I take a closer look at the paper Real-Time User-Guided Image Colorization with Learned Deep Priors, Zhang et al. (2017), and their colorization system. Based on this system, I will then introduce various ways in which to automatically choose color pixels in the original image, store them efficiently and assess their colorization performance, on the grayscale version, based on quality metrics and storage size. I found that the methods introduced can restore the original image color satisfactory with less than a kilobyte and almost perfectly with a few kilobytes of extra color data for 640x480 resolution. Nevertheless, it is still possible to improve the quality and reduce the size further with better point placement.</abstract>
    <identifier type="urn">urn:nbn:de:bvb:860-opus4-2988</identifier>
    <enrichment key="opus.source">publish</enrichment>
    <licence>Keine Creative Commons Lizenz (es gilt das deutsche Urheberrecht)</licence>
    <author>Daniel Ostertag</author>
    <subject>
      <language>deu</language>
      <type>uncontrolled</type>
      <value>Einfärbung</value>
    </subject>
    <subject>
      <language>deu</language>
      <type>uncontrolled</type>
      <value>Bildkompression</value>
    </subject>
    <subject>
      <language>deu</language>
      <type>swd</type>
      <value>Bildverarbeitung</value>
    </subject>
    <subject>
      <language>deu</language>
      <type>swd</type>
      <value>Farbenraum</value>
    </subject>
    <subject>
      <language>deu</language>
      <type>swd</type>
      <value>Maschinelles Lernen</value>
    </subject>
    <collection role="institutes" number="">Fakultät Informatik</collection>
    <thesisPublisher>Hochschule für Angewandte Wissenschaften Landshut</thesisPublisher>
    <thesisGrantor>Hochschule für Angewandte Wissenschaften Landshut</thesisGrantor>
    <file>https://opus4.kobv.de/opus4-haw-landshut/files/298/OstertagDaniel_BA.pdf</file>
  </doc>
</export-example>
