الفهرس 
introductionOnly 14 pages are availabe for public view
 |  | from | 86 |  |  |
| | | from | 86 | | |
Abstract
Games with a purpose (GWAP) and microtask crowdsourcing are considered two
techniques of the human-computation. Using these techniques can help in improving
the image retrieval systems (IMR) to be more accurate and helpful. It provides the
IMR system’s database with a rich of information by adding more descriptions and
annotations to images. One of the systems of human-computation is ESP Game. This
game is a type of games with a purpose. In the ESP game there has been a lot of work
were proposed to solve many of the problems in it and make the most benefit of the
game. One of these problems is that the ESP game neglects players’ answers for the
same image that don’t match. In this work three algorithms were proposed to solve
the problems of the ESP game. The first algorithm uses neglected data to generate
new labels for the images. This algorithm first focuses on measuring the total number
of labels generated by the proposed Recycle Unused Answers for Images algorithm
(RUAI). The RUAI algorithm was evaluated by a quality of labels measure. This
measure identifies the quality of the labels that were generated from the RUAI
compared to the pre-qualified labels from the ESP game dataset. The results reveal
that the proposed algorithm improved the results in compared to the ESP game in all
cases. The second algorithm help in generating informative ESP game labels with no
need to extra un-useful game rounds using one of the association rules mining
algorithms (FP-growth). The results show that new informative labels can be generated automatically without any interference of extra game rounds or any human.
The third algorithm was developed as a mobile game called MemoryLabel. It is a
single player mobile game. It helps in labeling images and gives description for them.
In addition, the game gives description for parts of the image not the whole image. In
addition, the game is published on Google play market for android applications. In
this trial, we first focused on measuring the total number of labels generated by our
game and also the number of objects that have been labeled. The results reveal that the
proposed game has promising results in describing images and objects. All three
algorithms were evaluated at the Menoufia University by the demonstrators in
computer science department.