Dr. Dobb's | Image Processing Trends

Image Processing Trends

In the past decade, the field of image processing (the processing of video and/or graphic images) and more specifically integrated circuits (ICs) for image processing, have changed dramatically

October 25, 2006
URL:http://www.drdobbs.com/embedded-systems/image-processing-trends/193402224

Introduction
In the past decade, the field of image processing (the processing of video and/or graphic images) and more specifically integrated circuits (ICs) for image processing, have changed dramatically.

The transition from SD to HD video content, the move to larger screen sizes and resolutions and the increasing use of compression technology has brought about new challenges and opportunities to merchant silicon providers.

Taking a Look Back
In the early 1990's the majority of flat panel TV (FPTV) manufacturing was taking place in Japan (Sharp, Sony, Panasonic, NEC, JVC, etc.) and these manufacturers were developing their display technologies and image processing solutions in-house.

At that time, manufacturers were able to take advantage of abundant internal fabrication (fab) capabilities, low tape-out costs and existing image processing solutions from other internal business units active in the flat panel computer monitor market.

Over time, the FPTV production base slowly made its way into Korea (LG & Samsung), as consumer demand for FPTV technology increased.. In many cases, in-house image processing solutions were not readily available and these manufacturers leveraged merchant image processing solutions developed primarily for the LCD computer monitor market.

During this same period, the LCD computer monitor market transitioned closer to a commodity space, encouraging many merchant monitor IC suppliers to changed course and enter the FPTV market. To be successful in the FPTV space, these merchant IC vendors needed to displace the widespread use of internal solutions. The strategy employed, was to offer a more 'cost effective' silicon solution through high levels of integration and 'good enough' image quality.

As the years progressed semiconductor processing technologies shrunk, fab NRE prices rose, and FPTV production spread across Asia. This increased the focus on better merchant image processing solutions. At that time At that time this merchant group consisted of Faroudja (latter acquired by Genesis), Genesis, Pixelworks, Silicon Image and Trident among others Faroudja dominated the pack with their DCDi technology, which is still regarded by some as the benchmark for SD image processing against which all other solutions are judged.. More recently Trident has gained strong momentum with a very good balance of integration, image quality and features.

Larger Screen Sizes and Higher Resolutions
In the early 2000's manufacturers began to introduce screen sizes greater than 50" and more recently introduced higher resolutions supporting full HDTV (1920x1080 pixels). These two factors, along with a lack of HDTV content made image artefacts (previously disguised by smaller sized displays and lower resolutions), much more apparent. This drove the need for better image processing quality!

To answer the call for better image quality, new entrants in the merchant semiconductor image processor space appeared such as Gennum offering VXP technology (migrating from their traditional broadcast market space, where the finest image quality is required), Silicon Optix with HQV, and Anchor Bay Technologies.

The main focus for this group of silicon providers was and is to provide better image quality and product differentiation through specialized functionality and processing.

The once broad image processing market has now been segmented. One group focuses on integration and lower cost (we call this the integration market) and the other group focuses on quality and differentiation, (we call this the specialization market).

What makes good image quality?
The simple answer is 'a lot of things' but there are certain areas of distinction such as de-interlacer performance, scaling performance, resolution (bit-depth), bandwidth, noise-reduction and frame rate conversion that sets image processors apart. Of course it goes without saying that there should be no compromise in image processing quality for both HD and SD content, as is the case with Gennum's VXP technology.

Creating an exceptional de-interlacer is somewhat like black magic. It requires a lot of time and effort to modify and tweak the algorithms over a number of silicon generations to get the right balance between sharpness and artefacts (jaggies, feathering, motion blur etc.). Figure 1 shows two images, one with proper pixel-by-pixel edge processing and one without. The difference is clearly visible!

Other areas for improvement include not only motion and edge adaptation but also speed adaptation. The concept is relatively straightforward, for video material with fast motion you want to employ a strict spatial interpolation method versus spatial-temporal processing for slower moving content. This minimizes temporal aliasing for faster moving content and preserves the maximum amount of detail for content with low or no motion.. Figure 2 shows speed adaptation in action on a vertically moving grid. With speed adaptive processing a sharp and clear grid is produced. Without speed adaptation the temporal alias can be seen as additional faint horizontal lines (in this case in the vertical direction).

The Latest Trend in Features
In recent years the trend has been to make improvements in the areas of image enhancement. More specifically, the areas of block artefact and mosquito noise reduction. These areas will increase in importance in the next few years as compression rates are turned up to squeeze as much video content as possible in to available distribution channels and onto media such as SD and HD discs. Figure 3 and 4 respectively highlight the possible improvements that can be achieved when employing the latest block and mosquito noise reduction algorithms.

Other typical areas of image enhancement improvements include adaptive contrast enhancement, sharpness and texture enhancement, and selective colour correction.

What Does the Future Hold?
The next big hurdle in the field of image processing will be the inclusion of motion compensated processing. As silicon providers' rush to incorporate new levels of processing (based on motion compensation) we will see a wide variety of offerings and various levels of quality and capabilities.

Initial implementations will focus on addressing frame rate conversions between 50 � 60Hz and from 60 � 120Hz. As demonstrated in Figure 5, significant gains can be achieved by using motion compensation techniques, but the challenge for silicon providers will be to find the optimal trade-off between cost (including things like die area, external memory requirements and off-chip bus bandwidth) and performance. A number of silicon generations will be required before silicon providers reach the optimal point.

As a manufacturer, what's right for you?
Your focus should be on specialization if you deem image quality as one of your core competencies and/or the key driver to differentiate your product and promote your brand.

What does this mean to the consumer?
The consumer will win! Until recently, innovation in merchant silicon image processors was slow to happen. With additional competition and technology advances consumers should see much shorter design cycles between new product offerings.

In the short-term, to take advantage of the new specialized offerings, the majority of consumers will need to purchase standalone scaler box products like the Crystalio II or Vantage-HD and future generations of these types of products.

Over the long-term, more manufacturers will begin to adopt specialized image processing across their product lines, offering the very best in image quality to consumers. As this model gains popularity, the need for standalone scaler boxes will be reduced and the advancements in technology will be made widely available at an affordable price.

Conclusion
With the increasing number of image processing merchant silicon providers, and the push for further refinements in image processing techniques, the end consumer will benefit the most due to advancing feature sets and lower prices.

As we move forward, manufacturers need to decide whether to focus on integration or specialization or both.

The future looks bright for the field of image processing and the consumer alike!

About the Author
David Vrhovnik is the Manager of Video Algorithm Development at Gennum Corporation of Burlington, Ontario, Canada. David obtained his Computer Engineering and Management degree from McMaster University (Hamilton, Ontario). He also has an MBA with a focus on Management of Innovation and Technology from the DeGroote School of Business (Hamilton, Ontario). If you have any comments or question about the article, you can contact him at [email protected] .