The Impact of Crowdsourcing on Image Similarity Identification using the Feature Detection Algorithms SIFT and SURF

  • warning: Parameter 1 to theme_field() expected to be a reference, value given in /srv/www/vhosts/groups/kdmrg/includes/theme.inc on line 170.
  • warning: Parameter 1 to theme_field() expected to be a reference, value given in /srv/www/vhosts/groups/kdmrg/includes/theme.inc on line 170.
  • warning: Parameter 1 to theme_field() expected to be a reference, value given in /srv/www/vhosts/groups/kdmrg/includes/theme.inc on line 170.
Present image search systems use the labeled texts alongside images to search for similar images. This comes with limitations such as failure to show images that are not properly labeled. Proper labeling of images in itself is not easy and has resulted to the use of crowds to solve this problem. Pioneering works such as Crowdsearch, and Sorokin have taken the initiative of proposing the use of images themselves to search for other similar images as opposed to the texts that describe these images. While these works in image recognition and search acknowledge the rising importance and capabilities of the crowd, none to our knowledge has made attempts to evaluate the interaction and impact of the crowd on the performance of existing image search algorithms. This work describes experiments that were carried out to explore this impact and the corresponding interaction between a crowd and two image recognition algorithms, SIFT and SURF. Preliminary results show that SURF performs better than SIFT while using a crowd in Amazon Mechanical Turk improves the identification of similar images of these two algorithms.