Another Word For It Patrick Durusau on Topic Maps and Semantic Diversity

March 23, 2013

Seeing the Future, 1/10 second at a time

Filed under: Image Understanding,Interface Research/Design,Usability,Users — Patrick Durusau @ 11:16 am

Ever caught a basketball? (Lot of basketball noise in the US right now.)

Or a baseball?

Played any other sport with a moving ball?

Your brain takes about 1/10 of a second to construct a perception of reality.

At 10 MPH, a ball moves 14.67 feet, while your brain creates a perception of its original location.

How did you catch the ball with your hands and not your face?

Mark Changizi has an answer to that question in: Why do we see illusions?.

The question Mark does not address: How does that relate to topic maps?

I can answer that with another question:

Does your topic map application communicate via telepathy or does it have an interface?

If you said it has an interface, understanding/experimenting with human perception is an avenue to create a useful and popular topic map interface.

You can also use the “works for our developers” approach but I wouldn’t recommend it.


About Mark Changizi:

Mark Changizi is a theoretical neurobiologist aiming to grasp the ultimate foundations underlying why we think, feel, and see as we do. His research focuses on “why” questions, and he has made important discoveries such as why we see in color, why we see illusions, why we have forward-facing eyes, why the brain is structured as it is, why animals have as many limbs and fingers as they do, why the dictionary is organized as it is, why fingers get pruney when wet, and how we acquired writing, language, and music.

January 22, 2013

Content-Based Image Retrieval at the End of the Early Years

Content-Based Image Retrieval at the End of the Early Years by Arnold W.M. Smeulders, Marcel Worring, Simone Santini, Amarnath Gupta, and Ramesh Jain. (Smeulders, A.W.M.; Worring, M.; Santini, S.; Gupta, A.; Jain, R.; , “Content-based image retrieval at the end of the early years,” Pattern Analysis and Machine Intelligence, IEEE Transactions on , vol.22, no.12, pp.1349-1380, Dec 2000
doi: 10.1109/34.895972)

Abstract:

Presents a review of 200 references in content-based image retrieval. The paper starts with discussing the working conditions of content-based retrieval: patterns of use, types of pictures, the role of semantics, and the sensory gap. Subsequent sections discuss computational steps for image retrieval systems. Step one of the review is image processing for retrieval sorted by color, texture, and local geometry. Features for retrieval are discussed next, sorted by: accumulative and global features, salient points, object and shape features, signs, and structural combinations thereof. Similarity of pictures and objects in pictures is reviewed for each of the feature types, in close connection to the types and means of feedback the user of the systems is capable of giving by interaction. We briefly discuss aspects of system engineering: databases, system architecture, and evaluation. In the concluding section, we present our view on: the driving force of the field, the heritage from computer vision, the influence on computer vision, the role of similarity and of interaction, the need for databases, the problem of evaluation, and the role of the semantic gap.

Excellent survey article from 2000 (not 2002 as per the Ostermann paper).

I think you will appreciate the treatment of the “semantic gap,” both in terms of its description as well as ways to address it.

If you are using annotated images in your topic map application, definitely a must read.

November 22, 2012

Developing New Ways to Search for Web Images

Developing New Ways to Search for Web Images by Shar Steed.

From the post:

Collections of photos, images, and videos are quickly coming to dominate the content available on the Web. Currently internet search engines rely on the text with which the images are labeled to return matches. But why is only text being used to search visual mediums? These labels can be unreliable, unhelpful and sometimes not available at all.

To solve this problem, scientists at Stanford and Princeton have been working to “create a new generation of visual search technologies.” Dr. Fei-Fei Li, a computer scientist at Stanford, has built the world’s largest visual database, containing more than 14 million labeled objects.

A system called ImageNet, applies the data gathered from the database to recognize similar, unlabeled objects with much greater accuracy than past algorithms.

A remarkable amount of material to work with, either via the API or downloading for your own hacking.

Another tool for assisting in the authoring of topic maps (or other content).

June 16, 2012

Does She or Doesn’t She?

Filed under: Image Processing,Image Understanding,Information Integration,Topic Maps — Patrick Durusau @ 2:57 pm

Information Processing: Adding a Touch of Color

From the post:

An innovative computer program brings color to grayscale images.

Creating a high-quality realistic color image from a grayscale picture can be challenging. Conventional methods typically require the user’s input, either by using a scribbling tool to color the image manually or by using a color transfer. Both options can result in poor colorization quality limited by the user’s degree of skill or the range of reference images available.

Alex Yong-Sang Chia at the A*STAR’s Institute for Infocomm Research and co-workers have now developed a computer program that utilizes the vast amount of imagery available on the internet to find suitable color matches for grayscale images. The program searches hundreds of thousands of online color images, cross-referencing their key features and objects in the foreground with those of grayscale pictures.

“We have developed a method that takes advantage of the plentiful supply of internet data to colorize gray photos,” Chia explains. “The user segments the image into separate major foreground objects and adds semantic labels naming these objects in the gray photo. Our program then scans the internet using these inputs for suitable object color matches.”

If you think about it for a moment, it appears that subject recognition in images is being performed here. As the researchers concede, its not 100% but then it doesn’t need to be. They have human users in the loop.

I wonder if the human users have to correct the coloration for an image more than once for a source of color image? That is does the system “remember” earlier choices?

The article doesn’t say so I will follow up with an email.

Keeping track of user-corrected subject recognition would create a bread crumb trail for other users confronted with the same images. (In other words, a topic map.)

December 27, 2011

Computer Vision & Math

Filed under: Image Recognition,Image Understanding,Mathematics — Patrick Durusau @ 7:10 pm

Computer Vision & Math

From the website:

The main part of this site is called Home of Math. It’s an online mathematics textbook that contains over 800 articles with over 2000 illustrations. The level varies from beginner to advanced.

Try our image analysis software. Pixcavator is a light-weight program intended for scientists and engineers who want to automate their image analysis tasks but lack a significant computing background. This image analysis software allows the analyst to concentrate on the science and lets us take care of the math.

If you create image analysis applications, consider Pixcavator SDK. It provides a simple tool for developing new image analysis software in a variety of fields. It allows the software developer to concentrate on the user’s needs instead of development of custom algorithms.

September 22, 2011

Explore large image collections with ImagePlot

Filed under: Image Understanding — Patrick Durusau @ 6:15 pm

Explore large image collections with ImagePlot from Flowing Data.

From the post:

When we make charts and graphs, we usually think of the data abstractions in terms of bars, dots, and other geometric shapes. ImagePlot, from UCSD-based Software Studies, instead makes it easier to use images to understand large collections.

Existing visualization tools show data as points, lines, and bars. ImagePlot’s visualizations shows the actual images in your collection. The images can be scaled to any size and organized in any order – according to their dates, content, visual characteristics, etc. Because digital video is just a set of individual still images, you can also use ImagePlot to explore patterns in films, animations, video games, and any other moving image data.

You can do this with other software (like R, for example), but ImagePlot is specifically built to handle lots of images (in the millions) and so it is much more robust, and it’s GUI-based, so no programming is required to use the software, which works on Windows, OS X, and Linux. The interface is pretty basic and not totally clear at first, but play around with the sample datasets and you should be able to pick it up fairly quickly.

The example I found particularly interesting was plotting Van Gogh painting by date on one axis and color on another.

A great deal of potential for exploring image collections from a variety of sources.

September 17, 2011

The Revolution(s) Are Being Televised

Filed under: Crowd Sourcing,Image Recognition,Image Understanding,Marketing — Patrick Durusau @ 8:17 pm

Revolutions usually mean human rights violations, lots of them.

Patrick Meier has a project to collect evidence of mass human rights violations in Syria.

See: Help Crowdsource Satellite Imagery Analysis for Syria: Building a Library of Evidence

Topic maps are an ideal solution to link objects in dated satellite images to eye witness accounts, captured military documents, ground photos, news accounts and other information.

I say that for two reasons:

First, with a topic map you can start from any linked object in a photo, a witness account, ground photo or news account and see all related evidence for that location. Granted that takes someone authoring that collation but it doesn’t have to be only one someone.

Second, topic maps offer parallel subject processing, which can distribute the authoring task in a crowd-sourced project, for instance. For example, I could be doing photo analysis and marking the location of military checkpoints. That would generate topics and associations for the geographic location, the type of installation, dates (from the photos), etc. Someone else could be interviewing witnesses and taking their testimony. As part of the processing of that testimony, another volunteer codes an approximate date and geographic location in connection with part of that testimony. Still another person is coding military orders by identified individuals for checkpoints that include the one in question. Associations between all these separately encoded bits of evidence, each unknown to the individual volunteers becomes a mouse-click away from coming to the attention of anyone reviewing the evidence. And determining responsibility.

The alternative, the one most commonly used, is to have an under-staffed international group piece together the best evidence it can from a sea of documents, photos, witness accounts, etc. An adequate job for the resources they have, but why settle for an “adequate” job when it can be done properly with 21st century technology?

GRASS: Geographic Resources Analysis Support System

GRASS: Geographic Resources Analysis Support System

The post about satellite imagery analysis for Syria made me curious about tools for use for automated analysis of satellite images.

From the webpage:

Commonly referred to as GRASS, this is free Geographic Information System (GIS) software used for geospatial data management and analysis, image processing, graphics/maps production, spatial modeling, and visualization. GRASS is currently used in academic and commercial settings around the world, as well as by many governmental agencies and environmental consulting companies. GRASS is an official project of the Open Source Geospatial Foundation.

You may also want to visit the Open Dragon project.

From the Open Dragon site:

Availability of good software for teaching Remote Sensing and GIS has always been a problem. Commercial software, no matter how good a discount is offered, remains expensive for a developing country, cannot be distributed to students, and may not be appropriate for education. Home-grown and university-sourced software lacks long-term support and the needed usability and robustness engineering.

The OpenDragon Project was established in the Department of Computer Engineering of KMUTT in December of 2004. The primary objective of this project is to develop, enhance, and maintain a high-quality, commercial-grade software package for remote sensing and GIS analysis that can be distributed free to educational organizations within Thailand. This package, OpenDragon, is based on the Version 5 of the commercial Dragon/ips® software developed and marketed by Goldin-Rudahl Systems, Inc.

As of 2010, Goldin-Rudahl Systems has agreed that the Open Dragon software, based on Dragon version 5, will be open source for non-commercial use. The software source code should be available on this server by early 2011.

And there is always the commercial side, if you have funding ArcGIS. The makers of ArcGIS, Esri support a several open source GIS projects.

The results of using these or other software packages can be tied to other information using topic maps.

October 25, 2010

Consensus of Ambiguity: Theory and Application of Active Learning for Biomedical Image Analysis

Consensus of Ambiguity: Theory and Application of Active Learning for Biomedical Image Analysis Authors: Scott Doyle, Anant Madabhushi Keywords:

Abstract:

Supervised classifiers require manually labeled training samples to classify unlabeled objects. Active Learning (AL) can be used to selectively label only “ambiguous” samples, ensuring that each labeled sample is maximally informative. This is invaluable in applications where manual labeling is expensive, as in medical images where annotation of specific pathologies or anatomical structures is usually only possible by an expert physician. Existing AL methods use a single definition of ambiguity, but there can be significant variation among individual methods. In this paper we present a consensus of ambiguity (CoA) approach to AL, where only samples which are consistently labeled as ambiguous across multiple AL schemes are selected for annotation. CoA-based AL uses fewer samples than Random Learning (RL) while exploiting the variance between individual AL schemes to efficiently label training sets for classifier training. We use a consensus ratio to determine the variance between AL methods, and the CoA approach is used to train classifiers for three different medical image datasets: 100 prostate histopathology images, 18 prostate DCE-MRI patient studies, and 9,000 breast histopathology regions of interest from 2 patients. We use a Probabilistic Boosting Tree (PBT) to classify each dataset as either cancer or non-cancer (prostate), or high or low grade cancer (breast). Trained is done using CoA-based AL, and is evaluated in terms of accuracy and area under the receiver operating characteristic curve (AUC). CoA training yielded between 0.01-0.05% greater performance than RL for the same training set size; approximately 5-10 more samples were required for RL to match the performance of CoA, suggesting that CoA is a more efficient training strategy.

The consensus of ambiguity (CoA) is trivially extensible to other image analysis. Intelligence photos anyone?

What intrigues me is extension of that approach to other types of data analysis.

Such as having multiple AL schemes process textual data and follow the CoA approach on what to bounce to experts for annotation.

Questions:

  1. What types of ambiguity would this approach miss?
  2. How would you apply this method to other data?
  3. How would you measure success/failure of application to other data?
  4. Design and apply this concept to specified data set. (project)

October 24, 2010

The Role of Sparse Data Representation in Semantic Image Understanding

Filed under: Bioinformatics,Biomedical,Image Understanding,Sparse Image Representation — Patrick Durusau @ 10:56 am

The Role of Sparse Data Representation in Semantic Image Understanding Author: Artur Przelaskowski Keywords: Computational intelligence, image understanding, sparse image representation, nonlinear approximation, semantic information theory

Abstract:

This paper discusses a concept of computational understanding of medical images in a context of computer-aided diagnosis. Fundamental research purpose was improved diagnosis of the cases, formulated by human experts. Designed methods of soft computing with extremely important role of: a) semantically sparse data representation, b) determined specific information, formally and experimentally, and c) computational intelligence approach were adjusted to the challenges of image-based diagnosis. Formalized description of image representation procedures was completed with exemplary results of chosen applications, used to explain formulated concepts, to make them more pragmatic and assure diagnostic usefulness. Target pathology was ontologically described, characterized by as stable as possible patterns, numerically described using semantic descriptors in sparse representation. Adjusting of possible source pathology to computational map of target pathology was fundamental issue of considered procedures. Computational understanding means: a) putting together extracted and numerically described content, b) recognition of diagnostic meaning of content objects and their common significance, and c) verification by comparative analysis with all accessible information and knowledge sources (patient record, medical lexicons, the newest communications, reference databases, etc.).

Interesting in its own right for image analysis in the important area of medical imaging but caught my eye for another reason.

Sparse data representation works for understanding images.

Would it work in other semantic domains?

Questions:

  1. What are the minimal clues that enable us to understand a particular text?
  2. Can we learn those clues before we encounter a particular text?
  3. Can we create clues for others to use when encountering a particular text?
  4. How would we identify the text for application of our clues?
« Newer Posts

Powered by WordPress