Perceptual metrics for image database navigation by Yossi Rubner

By Yossi Rubner

With the expanding variety of pictures to be had electronically, automated retrieval platforms have gotten crucial. This e-book introduces an absolute prerequisite for this sort of procedure: a metric, known as the Earth Mover's Distance (EMD), for evaluating photographs in phrases in their visual appeal. This metric describes the volume of labor that's essential to remodel one snapshot into one other, in a accurately outlined mathematical experience, and in a versatile and perceptually significant demeanour. a good linear programming set of rules allows the computation of this metric speedy sufficient for use for the interactive retrieval of pictures from huge repositories. The perceptual houses of the EMD, and the rate of its computation, bring about database navigation, a brand new paradigm for interacting with a repository of photographs. while navigating, the person is proven a really huge variety of pictures based on a question. The EMD among pairs of pictures, including a multidimensional scaling procedure, permits those photos to be displayed in order that comparable pictures seem with regards to one another on the pc display. during this means, the consumer can snatch at a look what is back, and will succeed in the photographs of curiosity with a small quantity of mouse clicks. A CD-ROM with complete colour photos is incorporated.

Extensive benchmark reviews and instance retrieval platforms convey the usefulness of the EMD and some great benefits of picture database navigation.

This ebook can be of curiosity to researchers, business execs, and graduate and post-graduate scholars within the fields of machine imaginative and prescient; photograph Processing; information Mining; electronic Libraries; Psychophysics; laptop technology; electric Engineering.

Show description

By Yossi Rubner

With the expanding variety of pictures to be had electronically, automated retrieval platforms have gotten crucial. This e-book introduces an absolute prerequisite for this sort of procedure: a metric, known as the Earth Mover's Distance (EMD), for evaluating photographs in phrases in their visual appeal. This metric describes the volume of labor that's essential to remodel one snapshot into one other, in a accurately outlined mathematical experience, and in a versatile and perceptually significant demeanour. a good linear programming set of rules allows the computation of this metric speedy sufficient for use for the interactive retrieval of pictures from huge repositories. The perceptual houses of the EMD, and the rate of its computation, bring about database navigation, a brand new paradigm for interacting with a repository of photographs. while navigating, the person is proven a really huge variety of pictures based on a question. The EMD among pairs of pictures, including a multidimensional scaling procedure, permits those photos to be displayed in order that comparable pictures seem with regards to one another on the pc display. during this means, the consumer can snatch at a look what is back, and will succeed in the photographs of curiosity with a small quantity of mouse clicks. A CD-ROM with complete colour photos is incorporated.

Extensive benchmark reviews and instance retrieval platforms convey the usefulness of the EMD and some great benefits of picture database navigation.

This ebook can be of curiosity to researchers, business execs, and graduate and post-graduate scholars within the fields of machine imaginative and prescient; photograph Processing; information Mining; electronic Libraries; Psychophysics; laptop technology; electric Engineering.

Show description

Read or Download Perceptual metrics for image database navigation PDF

Best organization and data processing books

JDBC Recipes: A Problem-Solution Approach

JDBC Recipes presents easy-to-implement, usable strategies to difficulties in relational databases that use JDBC. it is possible for you to to combine those options into your web-based purposes, comparable to Java servlets, JavaServer Pages, and Java server-side frameworks. this useful publication lets you minimize and paste the recommendations with none code alterations.

The effects of sterilization methods on plastics and elastomers: the definitive user's guide and databook

This largely up to date moment version used to be created for scientific machine, scientific packaging, and meals packaging layout engineers, fabric product technical aid, and research/development team of workers. This accomplished databook comprises vital features and homes info at the results of sterilization tools on plastics and elastomers.

Additional resources for Perceptual metrics for image database navigation

Sample text

Red and green) [103]. 2. Perceptually Uniform Color Spaces For an image retrieval system, it is important to be able to measure differences between colors in a way that matches perceptual similarity as well as possible. This task is simplified by the use of perceptually uniform color spaces. A color space is perceptually uniform if a small perturbation of a color will produce the same change in perception anywhere in the color space. In 1976, the CIE standardized two perceptually uniform color spaces (since no single system could be agreed upon), Ä£ Ù£ Ú £ (CIELuv) and Ä£ £ £ (CIELab).

Intuitively, the solution is the minimum amount of “work” required to transform one signature into the other. ” In Section 1 we formally define the EMD. We discuss its properties in Section 2 and specify computational issues, including lower bounds, in Section 3. In Section 4 we describe a saturated ground distance and claim better agreement with psychophysics than simpler distance measures. Finally, in Section 5 we mention some extensions to the EMD. 15 The Earth Mover’s Distance 1. DEFINITION The EMD is based on the following linear programming problem: Let ´Ô½ ÛÔ½ µ ´ÔÑ ÛÔÑ µ be the first signature with Ñ clusters, where Ô is the cluster representative and ÛÔ is the weight of the cluster; É ´Õ½ ÛÕ½ µ ´ÕÒ ÛÕÒ µ the second signature with Ò clusters; and the ground distance matrix where ´Ô Õ µ is the ground distance between clusters Ô and Õ .

For example, consider the following two color images: In the first, there are blue skies above a green field, while in the other there is a blue lake below green plants. Although the color distributions might be very similar, the positions of the colors in the image are very different and may have to be taken into account by the query. This can be achieved by modifying the color distance in Section 3 as follows: Instead of using the three-dimensional CIELab color space, we use a five-dimensional space whose first three dimensions are the CIELab color space, and the other two are the ´Ü Ý µ position of each pixel.

Download PDF sample

Rated 4.36 of 5 – based on 22 votes