Gestalt Perception For Decomposition of Images

The objective of this project is to conduct experiment on images to decide which is the ground truth parse for humans to decompose elements of those images in first glacnce. The method of perceiving an image and recognising patterns by human vision is explained by Gestalt Theory. The theme of Gestalt theory is "The whole is more than the sum of their parts". Gestalt Theory is widely used in grouping and figure recognition in vision in Artificial Intelligence.

The main motivation for doing this experiment came from the fact that Gestalt Principles are widely implemented by computation models for object segmentation from a given image. Like in paper [2], the author introduces a methodology to learn relations inferred from Gestalt principles and an application to segment unknown objects, even if the objects are stacked or jumbled and tackle the problem of segmenting partially occluded objects.

The experiments will be done using eye gaze tracking system. Subjects will be shown images which they have to decompose in their mind. Their saccadic eye movements will be traced by the gaze tracking system. The time duration for a single image will be very small( about 3-5 seconds). The image set for the experiments will be generated using Microsoft Paint/Word. Some of the example images are shown below.

Later each subjects will be asked what did they decompose in a perticular image and then the answer will be compared to the tracked image. Finally I will try to conclude the result of the experiment.

1. Dejan Todorovic (2008) Gestalt principles. Scholarpedia, 3(12):5345.
2. Implementation of Gestalt Principles for Object Segmentation, Andreas Richtsfeld, Michael Zillich and Markus Vincze, Automation and COntrol Institute(ACIN), Vienna University of Technology, 2012
3. A Century of Gestalt Psychology in Visual Perception: I. Perceptual Grouping and Figure–Ground Organization. -Wagemans, Johan and Elder, James H and Kubovy, Michael and Palmer, Stephen E and Peterson, Mary A and Singh, Manish and von der Heydt, 2012