According to industry sources, over 1.1 billion mobile phones were sold in 2008, and about 80% of them had cameras. This boom in camera phone adoption has allowed camera phone sales to surpass traditional camera sales. According to Research and Markets (Ireland), the market for camera phones will be segmented into three major categories:
* High-end: smart phones with open platforms (those supporting third party applications or using open source operating systems) and uncompromised multimedia and data capabilities
* Mid-range: feature phones with (generally) closed platforms. However, these phones are upgradeable and customizable systems and feature very good multimedia and data capabilities
* Low-end: low-end phones with closed systems and limited multimedia and data capabilities
Each segment is expected to be further sub-divided into “micro-segments” that support a range of camera features (video capture, high-resolution, zoom, low-light performance, etc.). The result is a wide-range of cameras for many applications in almost a billion camera-phones.
This enormous growth of in camera phones has created a host of design problems for phone manufacturers, not the least of which is image sensor calibration. For example, many engineers that are asked to integrate cameras into phone designs do not have extensive academic training in image science.
While the hardware design is relatively straightforward, tuning the visual performance of the sensor is much more complex. Most sensors have a bank of registers that must be properly set in order to get a viable image and/or video stream. These registries are inter-dependent, and setting or adjusting them can be a complex undertaking due to these interdependencies.
To get a desired image, it is necessary to tweak certain register settings, and because of these tweaks, other registers will likewise need to be adjusted. Furthermore, the settings required to produce a viable image will vary from image sensor to image sensor, even among sensors manufactured by the same company.
Although many sensor vendors provide default register values, these default settings often do not address the variations in image performance required by different customers.
For example, Customer A and Customer B may have a different opinion on what the color red should look like. Customer A may want a red with more blue tones, while Customer B may prefer a red with more orange tones. The difficulty in calibrating an image sensor becomes readily apparent.
Currently, many designers conduct subjective and/or objective evaluations to determine if their cameras are delivering the required image quality. Subjective evaluation is based on human perception on image quality by viewing the image alone. Companies using the subjective evaluation often use the eyes of an imaging expert or a panel of experts to determine if image quality is acceptable for the target application and/or market.
This approach is valid in the sense that it is closest to what the end user will experience. However, subjective evaluation can be difficult to duplicate, especially if there is turnover among a company's resident image expert(s), which can lead to a gradual change in what a company views to be a quality image.
To counter the inconsistencies that can result from subjective valuation, many companies use quantitative evaluation to supplement or replace subjective evaluation. Quantitative evaluation subjects the camera to repeatable test conditions to measure items such as signal-to-noise ratio, well capacitance, sensitivity, color accuracy and sharpness.
Software tools available industry-wide, the Imatest utility, for example, have been developed to facilitate quantitative evaluation. However, these tools are only used to test image quality; they do not offer any ability to tune the sensor in real time with quantitative scores.
To assist customers with the calibration of their image sensors, some sensor vendors are now offering software packages designed to simplify the work of image calibration/tuning.
However, vendors develop these tools to work exclusively with their image sensor products and vendors' tools can vary greatly in quality and ease-of-use. For a designer unfamiliar with image sensor calibration, it can be difficult to determine if a vendor's image sensor utility will provide the required functionality.
In light of this, a quick review of the image sensor specifications most important to creating a quality image would be useful. In order for a vendor calibration tool to be useful, it should allow easy modification /tuning of the following:
* Image pipeline – this refers to the image processing blocks built in the sensor. A useful calibration tool should be able to illustrate how the registers are grouped to control each block and the correlations between them.
* Lens shading correction -The lens shading refers to the fall-off of light intensity across the image height due to the change in the incident light angle. The correction is often a built in function of the sensor.
It applies one or multiple reverse gain curve against the fall-off to the pixel array. A good calibration tool should be able to check the uniformity in each color plane and the blemish level caused by the gain change to the pixel array, and to make the trade-off between the correction level and negative impact.
* Color reproduction – designers need to be able to easily tune a sensor's color matrix settings to meet the color accuracy or to match a reference color template. Additionally, designers will need to adjust color settings to respond to differing light levels/conditions to maintain accurate color reproduction.
* Sharpness – Sharpness has a direct impact on visual perception. Tuning sharpness using real time SFR measurement results is an effective approach. In addition to lens characteristics, noise reduction and edge enhancement also have a direct impact on sharpness. They are subject to tune with SNR and other parameters.
* SNR – Signal to noise ratio performance is one of the sensor's key characteristics and effects sharpness, color and other portions of the image pipeline. Because of this, it has to be carefully monitored during sensor tuning.
To further facilitate ease-of-use for camera phone designers, a vendor's image sensor calibration utility should use a GUI with a short learning curve. Vendor-provided tools like this will allow camera phone designers to bypass the “best guess” and “trial-and-error” approaches to image sensor calibration and let them focus on their main goal: bringing exciting new camera phones to market quickly and efficiently.
John Lin, Sr. is Design Engineering Manager; Jean Chao is Business Development Manager; and Shri Sundaram is Business Development Manager at Toshiba America Electronic Components, Inc.