Interactive Co-segmentation of Objects in Image Collections by Dhruv Batra, Adarsh Kowdle, Devi Parikh, Jiebo Luo, Tsuhan

By Dhruv Batra, Adarsh Kowdle, Devi Parikh, Jiebo Luo, Tsuhan Chen (auth.)


Show description

Read or Download Interactive Co-segmentation of Objects in Image Collections PDF

Similar graphics & multimedia books

Remote Sensing of the Environment and Radiation Transfer: An Introductory Survey

The interplay of the sunlight and warmth radiation with the ambience and floor is the topic of the booklet. it truly is important additionally for broad circle scientists excited by environmental reports. The ebook comprises the outline of 17 laptop learning courses assisting assorted subject matters of classes. It contains purely the bottom floor for comprehension of key themes and gives the accomplishment of functional works with utilizing in particular elaborated desktop courses.

Emerging Trends in Computing and Communication: ETCC 2014, March 22-23, 2014

The booklet provides papers introduced through researchers, commercial specialists and academicians on the convention on rising tendencies in Computing and conversation (ETCC 2014). As such, the ebook is a set of contemporary and leading edge works within the box community defense and Cryptography, Cloud Computing and massive info Analytics, information Mining and information Warehouse, communique and Nanotechnology and VLSI and photograph Processing.

Guide to Signals and Patterns in Image Processing: Foundations, Methods and Applications

This article experiences the sector of electronic photo processing from the several views provided through the separate domain names of sign processing and trend popularity. The e-book describes a wealthy array of purposes, representing the most recent tendencies in and educational learn. To motivate additional curiosity within the box, a variety of worked-out numerical difficulties can be integrated within the textual content.

Location-Aware Applications

Cellular consumers wish leisure, enterprise apps, and on-the-go companies that realize and reply to situation. This booklet will consultant you thru the know-how and enterprise of cellular purposes so that you can create aggressive and cutting edge apps in keeping with location-based providers. it really is an attractive examine the LBS panorama, from selecting the best cellular platform, to creating cash together with your program, to facing privateness concerns.

Extra info for Interactive Co-segmentation of Objects in Image Collections

Example text

For example, gamers want to be able to “scan” and use their own gear (like skateboards) in a skateboarding game. The computer-vision task then is to create a 3D model of this object of interest from a collection of images of this object. One approach to achieve this, would be to haul an expensive laser scanner to get precise depth estimates in a controlled setup, and reconstruct the object [16]. However, this might not be a feasible solution for average users. Another typical approach is to capture images of the object in a controlled environment like a multicamera studio with mono-color screen [1, 3, 6, 8, 22, 24] or structured lighting [25], and then use something like a shape-from-silhouette algorithm [2, 5, 7, 23] to construct the 3D model.

We extract the following features to describe an image. • Illumination histogram: One of the observations we made was that the presence of strong shadows across the image (Fig. 10(a)) often results in poor co-segmentation accuracies. To capture this intuition, we compute a 50-dim histogram of the gray scale image. • Hue, Saturation and Value entropy: The variety of colours in an image typically corresponds to the amount of useful information in the image, as seen in Fig. 10(b). We quantify this via a 3-dim vector holding the entropies of the hue, saturation and value marginal histograms.

Fig. 18: Common Failure Case: (a) shows the group of images; (c) shows the segmentations achieved by scribbling on a single image, shown in (b). Cyan indicates foreground. (f) shows the segmentations achieved by scribbles on multiple images, shown in (e). When foreground and background have a lot of overlap in colour distributions, our interactive segmentation method faces difficulty in producing accurate segmentations (compare segmentations in (c),(f) with ground-truth in (d)). g. colour-pallet of [12]), which should result in better performance.

Download PDF sample

Rated 4.35 of 5 – based on 46 votes