This dissertation was presented to the Faculty of the Graduate School of The University of Texas at Austin in partial fulfillment of the requirements for the degree of

Ph.D. in Electrical Engineering


Composition Guided Image Acquisition  


Serene Banerjee, Ph.D.E.E.

The University of Texas at Austin, August 2004


Prof. Brian L. Evans


Defense Slides - Dissertation - Images in Dissertation


To make a picture more appealing, professional photographers apply a wealth of photographic composition rules, of which amateur photographers are often unaware. This dissertation aims at providing in-camera feedback to the amateur photographer while taking pictures. The proposed algorithms do not depend on prior knowledge of the indoor/outdoor setting or scene, and are amenable to software implementation on fixed-point programmable digital signal processors available in digital still cameras.

The key enabling step in automating photographic composition rules is to locate the main subject. Digital still image acquisition maps the 3-D world onto a 2-D picture. Using the 2-D picture alone, segmenting the main subject without prior knowledge of the scene is ill-posed. Even with prior knowledge, segmentation is often computationally intensive and error prone.

This dissertation defends the idea that reliable main subject segmentation without prior knowledge of scene and setting may be achieved by acquiring a single picture, in which the optical system blurs objects not in the plane of focus. After segmentation, photographic composition rules may be automated. In this context, segmentation only needs to approximately and not precisely locate the main subject.

In this dissertation, I combine optical and digital image processing to perform the segmentation of the main subject without prior knowledge of the scene. In particular, I propose to acquire a picture in which the main subject is in focus, and the shutter aperture is fully open. The lens optics will blur any object not in the plane of focus. For the acquired picture, I develop a computationally simple one-pass algorithm to segment the main subject.

The post segmentation objective is to automate selected photographic composition rules. The algorithms can either be applied on the picture taken with the objects not in the plane of focus blurred, or on a user-intended picture with the same focal length settings. This way, in-camera feedback can be provided to the amateur photographer, in the form of alternate compositions of the same scene.

I automate three photographic composition rules:

  1. placement of the main subject obeying the rule-of-thirds,
  2. background blurring to simulate the main subject being in motion or decrease the depth-of-field of the picture, and
  3. merger detection and mitigation when equally focused main subject and background objects merge as one object.

The primary contributions of the dissertation are in digital still image processing. The first is the automation of segmentation of the main subject in a single still picture assisted by optical pre-processing. The second is the automation of main subject placement, artistic background blur, and merger detection and mitigation to try to improve photographic composition.


For more information contact: Serene Banerjee <>