Saliency as compact regions for local image enhancement
The standard workflow to increase the perceived salience of an image's main subject consists in identifying the region of interest, and processing that region according to a set of rules. The level of analysis and processing can greatly vary, from increasing saturation or sharpness to identifying semantic concepts, e.g., faces, and employ a complex, tailored, modification.
This is a delicate problem to approach: saliency prediction algorithms are currently not precise enough, and region classification is necessarily limited to a few specific classes. Furthermore, the variety of content often precludes the usage of a fixed set of rules in the enhancement step.
Rather than attempting to predict saliency in images, we propose that important regions are somewhat distinct from their surroundings and can be identified by features that are spatially compact, in addition to standard compositional cues. Having identified the region of interest, we provide an enhanced image by increasing the values of its compact feature(s), i.e., increasing the perceived saliency of the region of interest. Preference studies indicate our modified images are significantly preferred to the original ones.
Document Type: Research Article
Publication date: 2011-01-01
CIC is the premier annual technical gathering for scientists, technologists, and engineers working in the areas of color science and systems, and their application to color imaging. Participants represent disciplines ranging from psychophysics, optical physics, image processing, color science to graphic arts, systems engineering, and hardware and software development. While a broad mix of professional interests is the hallmark of these conferences, the focus is color. CICs traditionally offer two days of short courses followed by three days of technical sessions that include three keynotes, an evening lecture, a vibrant interactive (poster) papers session, and workshops. An endearing symbol of the meeting is the Cactus Award, given each year to the author(s) of the best interactive paper; there are also Best Paper and Best Student Paper awards.
Please note: for Purposes of its Digital Library content, IS&T defines Open Access as papers that will be downloadable in their entirety for free in perpetuity. Copyright restrictions on papers vary; see individual paper for details.
- Information for Authors
- Submit a Paper
- Subscribe to this Title
- Membership Information
- Terms & Conditions
- Ingenta Connect is not responsible for the content or availability of external websites