Friday, January 28, 2011

EI papers

The SPIE/IS&T Electronic Imaging Symposium is over and being wrapped up. The event was successful and the increased attendance is a positive signal of the improving economy. The quality of the papers was generally high and the technical discussions in the corridors were intense, letting us hope for good papers being incubated from this synergistic interactions and being presented at EI 2012 in the same location 22–26 January.

Our papers were very well received. We received many questions on our paper "Is it turquoise + fuchsia = purple or is it turquoise + fuchsia = blue?", which was generally referred to as "the calibrated lunch paper." The paper is now available online at this permalink: http://dx.doi.org/10.1117/12.872581.

This year's traditional session on the Dark Side of Color will forever be remembered for Dr. Gabriel Marcu interrupting the presentation as Dr. Bob Ulichney was sneaking out of the room and announcing "Dr. Ulichney has left the building" in reference to Elvis.

Our paper for the session on the Dark Side of Color was more about being in the dark on color. The title "ICC profiles: are we better off without them?" is of course very controversial, but when the consequences are in the billions of dollars, the gloves are off and print service providers (PSP) will do what they have to do. In the hallways, this paper was referred to by a comment made in the Q&A period: "is it legal to profile ICC profiles?" The paper is now available online at this permalink: http://dx.doi.org/10.1117/12.880240.

I hope next year we can counter this question with plenty of detailed information. One thing though will be sure: the committee dinner will be again at Sakae.

Tuesday, January 25, 2011

ICC profiles: are we better off without them?

In ten minutes we are presenting our provocative paper on ICC profiles in the session on the Dark Side of Color in the Harbour Room A here at EI. In case you did not make it, here are our slides:



Is it turquoise + fuchsia = purple or is it turquoise + fuchsia = blue?

In five minutes we are presenting our paper on arithmetic with color terms in the Harbour Room A here at EI. In case you did not make it, here are our slides:



Here are the speaker's notes:

{Color diagrams by scientists}

When color scientists draw chromaticity diagrams to illustrate their publications, they do it in black and white. Color, at most, is used for annotation purposes, not as self-reference.

{Color diagrams by designers}

This frustrates color designers, who make ample use of colors and expect more guidance from color scientists using the designer's language. Therefore, they prefer colorized chromaticity diagrams and they like how they can pencil in color theories because chromaticity diagrams can be used to predict the mixture of aperture color.

However, when they explain color theories using color terms, they get confused, because they have samples for the basic color terms and use them to label the colorized chromaticity diagram.

{sRGB color diagram}

In fact, the colorization of chromaticity diagrams is just that: a colorization. It bears little relation to the color's self-reference, because there is not device gamut covering the entire chromaticity diagram.

This figure shows how a chromaticity diagram can be more accurately colorized on an sRGB device.

{It is 3-dimensional!}

In the yellow range, the colorization fails and sticks out of the chromaticity diagram. This is because the gamut is tridimensional and the gamut rendering algorithm is somewhat imprecise. On the right side we move the viewpoint.

{Printer gamut}

You are seeing the projected image from a beamer, which in reality is not an sRGB device, so right now you are not seeing the correct self-referenced colors. In the case of a printer the error is even larger, but if for a moment you imagine viewing the diagram at right on a SWOP printer, you realized how off the colorized chromaticity diagram we saw earlier is.

{Color theories — an abstraction}

From a color science point of view, the various color theories have their own problems. Compare for example Leonardo's color circle with that proposed by Itten. The color are quite different. What happened?

{Itten revealed}

Of Leonardo we know that he considered color to be a perceived entity and his theory is based on color opponency. We discover Itten's method by looking at the black and white version of the color circle published in the student version of his book: He simply started with monolexemic color names and then interpolated with bilexemic color names to obtain a richer circle.

{Color by measurement}

We conclude, that we cannot predict arbitrary color mixtures by doing arithmetic with color terms obtained from colorized chromaticity diagrams. The general way to predict color mixtures is to reproduce actual samples and then measure them.

{Naming color}

How then can we determine the proper color terms? The World Color Survey has shown how to do it for the colors on the surface of the Munsell color gamut, and Boynton and Olson have shown how to do it for the full OSA color gamut.

Both experiment classes are limited to the basic color terms. For a large number of color terms, we need data from thousands of observers, which we can hardly do using controlled psychophysics techniques. Possible approaches include Amazon's Mechanical Turk and open Web surveys based on crowd-sourcing.

{Validating crowd-sourcing}

Unfortunately, non-withstanding Boynton's positive assessment of the uncontrolled protocol used in the World Color Survey, crowd-sourcing is still considered very controversial in the color science community. Therefore, we decided to perform a controlled experiment with an unusually large number of observers.

{Color chips given and color seating arrangement}

We printed a number of color chips and gave them to the researchers having lunch at the cafeteria and asked them to sit at the labelled tables with the term closest to their color chip.

{Calibrated lunch vs. Web color centroids}

As you can see from this diagram, we achieved excellent correlation between the controlled lunch room experiment and the uncontrolled crowd-sourced experiment on the Web. Thus, our data from the Web is scientifically valid.

{Beyond crowd-sourcing}

One question arising when teaching color design is which synonym to use for a given color. To answer this question we need more data than even crowd-sourcing can provide. Also, we have repeatedly written how color terms are ephemeral and change with time.

In this diagram we used Google's English book corpus and tracked the number times fuchsia and magenta appear in book each year from 1800 to 2000. The appearance of the term magenta coincides with the historical facts and the numbers suggest your should prefer the term magenta over fuchsia.

{What is the best term for RGB = (0, 1, 1)?}

This diagram illustrates how cyan is not a good choice and you should use the term turquoise instead.

{Is the term for RGB = (1, 1, 0) monolexemic?}

We can use this approach also to answer very difficult questions, like the number of basic color terms. This diagram illustrates that olive green might be becoming a basic term also in English, because chartreuse is monolexemic.

Our data shows that few people ever learned to correctly spell fuchsia, but the data shows this is not critical. However, you should definitely learn how to spell chartreuse.

Thursday, January 20, 2011

Warmed-up Color Slides

Around 1986, I prepared a set on slides on digital color reproduction. In 1990, enough people bugged me about them, that I converted them from Tioga to FrameMaker. Over the Nineties, I gave this tutorial—which at some point I renamed to Understanding Color because it evolved more into an introduction to color science than a primer in color reproduction—many times, each time updating it with the latest feedback.

In 2004, a benefactor gave me a laptop and I converted the slides from portrait format for overhead projectors to landscape format for beamers. In 2008 the benefactor surprised me with a new laptop, for which there no longer was FrameMaker, so once again I converted the slides, this time to LaTeX. The old Frame slides, I made available on SlideShare.

As a society, in this country we are now investing our wealth more in the financial industry than in technology. If I look at the PISA scores, I should move from Palo Alto to Shanghai, but I guess I am too old now for switching to a fifth culture. Therefore, since the slides were viewed, downloaded, embedded by more than 3000 people, I am making available the 2010 version for the benefit of whoever can make good use of them.

Wednesday, January 19, 2011

Parallel Processing for Image Recognition

In a few days, imaging technologists from around the world will be flocking to the San Francisco Airport Hyatt to attend the Electronic Imaging Symposium.

Monday 24 January from 10:40 AM to 11:10 AM many delegates will fasten their seat-belts in Sandpebble Room D, where IS&T Fellow and HP Labs Director and Distinguished Technologist Dr. Steven J. Simske will be giving his Invited Talk on Parallel Processing Considerations for Image Recognition Tasks in the Conference on Parallel Processing for Imaging Applications.

Many image recognition tasks are well-suited to parallel processing. The most obvious example is that many imaging tasks require the analysis of multiple images. From this standpoint, then, parallel processing need be no more complicated than assigning individual images to individual processors. However, there are three less trivial categories of parallel processing that will be considered in this paper: parallel processing (1) by task; (2) by image region; and (3) by meta-algorithm.

Parallel processing by task allows the assignment of multiple workflows—as diverse as optical character recognition [OCR], document classification and barcode reading—to parallel pipelines. This can substantially decrease time to completion for the document tasks. For this approach, each parallel pipeline is generally performing a different task. Parallel processing by image region allows a larger imaging task to be sub-divided into a set of parallel pipelines, each performing the same task but on a different data set. This type of image analysis is readily addressed by a map-reduce approach. Examples include document skew detection and multiple face detection and tracking. Finally, parallel processing by meta-algorithm allows different algorithms to be deployed on the same image simultaneously. This approach may result in improved accuracy.

Useful links:

Monday, January 17, 2011

Parallel Transparency

Technology allows everybody to do their own work without assistance. When office automation software programs allowed office workers to create professional quality documents, graphic artists had to take the sophistication of high-concept design up to the next level, above the abilities of office tools.

One of the key techniques has been the heavy usage of transparency. Consequently, commercial printers see a large number of documents containing transparency. The specification of transparency in PDF is very sophisticated, well above to the simple transparency used for example in video games.

Therefore, adding transparency to a GPU-based RIP is quite a challenging task. Indeed, not only has the complex PDF transparency to be implemented, but it also necessary to implement an ICC color management module on the GPU. And it all has to work on tiled images.

At the Electronic Imaging Symposium, John Ludd Recker from HP Labs will report on his experience implementing GPU-based transparency in Ghostscript. His lecture on A GPU accelerated PDF transparency engine will be in the Conference on Parallel Processing for Imaging Applications.

Useful links:

Friday, January 14, 2011

Parallel Error Diffusion

From the earliest days of digital color reproduction, there has been a need to add vector processing units to achieve viable executions times. For many imaging operations, algorithms can easily be vectorized because they operate independently on the pixels. However, some operations are spatial: sharpening, compression, error diffusion halftoning, etc.

Thursday, January 13, 2011

Why Being a Nerd Is Hazardous to Your Health

In a steady stream of recent papers, social psychologists have identified several potentially unhealthy changes in the cardiovascular, immune, and nervous systems of chronically lonely people. The findings could help explain why epidemiological studies have often found that socially isolated people have shorter life spans and increased risk of a host of health problems, including infections, heart disease, and depression. The work also adds a new wrinkle, suggesting that it's the subjective experience of loneliness that's harmful, not the actual number of social contacts a person has. An impressive network of collaborations with researchers in other disciplines is now pioneering a new science of loneliness.

Read the story in Science 14 January 2011: Vol. 331 no. 6014 pp. 138-140

GPU-Completeness

In the last decade, the computing industry has undergone a major revolution: software from suites to apps and hardware from workstations to mobile devices. The hardware platform of choice is no longer a souped up high clock-rate server class microprocessor but a system on a chip (SoC) combining a CPU, GPU, memory controller, etc., all running at very low power. The challenge is then how to partition a computation between CPU and GPU, which is done at runtime through an OpenCL kernel. This kernel is very difficult to write, because today it is based on arcane heuristics, not on an algorithm based on the system's current state.

At the Electronic Imaging Symposium, Dr. I-Jong Lin from HP Labs will present a new theory to solve this problem. His lecture on GPU-Completeness: Theory and Implications will be in the Conference on Parallel Processing for Imaging Applications. In a nutshell, when an algorithm is transformed from serial to parallel, there is a loss of accuracy. When the accuracy vs. parallelism trade-off is exponential rather than polynomial, the problem belongs in the class of GPU-Completeness.

The algorithmic class of GPU-Completeness is defined in terms of input, output, parallel performance, and a quality metric. Dr. Lin will validate his theory with experimental data from imaging applications: color transformation, halftoning, and run length encoding.

Useful links:

Saturday, January 8, 2011

The Silencing of the Lambdas

Silencing is a newly observed optical illusion that reveals how difficult it is for the brain's visual system to notice when moving objects change some of their properties by a relatively small amount (λ).

To better understand the significance, see these animated demos.

Tuesday, January 4, 2011

Register today for EI

Today is the last day to save $100 over onsite pricing for the Electronic Imaging symposium registration.