High-resolution cameras offer more pixels, but there are the downsides to cramming
more pixels on limited amounts of silicon. Alf Chang, a&s consultant and a former
installer, looks at how component selection affects image quality.
High-resolution cameras offer more pixels, but there are the downsides to cramming more pixels on limited amounts of silicon. Alf Chang, a&s consultant and a former installer, looks at how component selection affects image quality.
Component selection directly affects camera performance. A&S recently looked at 19 megapixel cameras in a controlled test environment to find out how image quality is affected by component differences as well as product design. We wanted to uncover overlooked or hidden issues, as well as key areas for users and installers to watch for. Our test was designed for maximum fairness and shown to objective judges. We positioned the cameras in a circle, closed off at the top and bottom like a drum. Inside the drum, the cameras monitored the same scene of a train track and figurines, lit with controlled illumination connected to a lux meter.
One factor we looked at was the relationship between image sensor size and the number of pixels. The 19 camera models surveyed used various image sensor formats from different providers. It is clear that going from standard definition (SD) to megapixel resolution affects the size of the image sensor. Nearly all 1/3-inch CMOS sensors were 1.3-megapixel — either 1,280 by 720 pixels (720p) or 1,280 by 960 pixels. For 1/2-inch sensors, most of them were 2-megapixel or 3-megapixel, although a few models went up to 5 or even 10 megapixels. Thus, we wondered about the size of the image sensor and the number of pixels. Based on our tests, 1/3-inch CMOS sensors should have 1.3-megapixel resolution on the silicon, representing 1.3 million individual pixels. However, the actual size of the 1/3-inch sensor was nearly the same as a D1 or VGA sensor with 350,000 pixels (720 by 480). On the same 1/3-inch CMOS sensor, a pixel for a D1 sensor is about 7.5 nanometers, but shrinks to 3.5 nanometers for a 1.3-megapixel sensor. The 1/3-inch megapixel CMOS sensor will have greater detail and resolution than a D1 sensor. However, a 1/2-inch megapixel sensor will have larger pixels, resulting in better light sensitivity. At the same time, the effective resolution of CCDs tops out at 2 megapixels. The CCD imaging process is slower than CMOS, which is not ideal for high-resolution output. From this test, we can see that network cameras are limited by CCDs for increased resolution.
In our tests, we found a highly unusual feature of the image sensors. During camera assembly, image sensors can be misaligned — when the surface of the sensor is not perfectly flat. This results in the edges of the image — the top, bottom, left or right — to look blurred, as seen in Image A's top right and bottom right corners. This is atypical, but happens more often than one would think.
Inexperienced installers might think the misaligned image sensor was a lens issue and constantly adjust the focus. In this test, we hit on a previously unknown secret. In selecting megapixel surveillance cameras, component selection has important implications on image quality.
SAME SENSORS, DIFFERENT RESULTS
The same sensor paired with the same decoding chip can produce completely different images. Each camera manufacturer has its own selection criteria for components. Apart from the image sensor, there is also the compression SoC, image sensor processor (ISP) for image correction and power supply components. While most cameras look nearly the same under the hood, our test showed the decoding SoC plays a crucial role.
The cameras we tested are similar on the inside. Depending on local preferences, camera manufacturers selected specific SoC brands. In Table 2, we can see the different SoCs represented by the 19 cameras, which are mostly new models from top brands. Thus, the SoCs selected are from reputable and more expensive brands. Different cameras using the same image sensor, decoder and ISP produced visibly distinct results in color performance. Even with the same components, the cameras looked markedly different for back-light compensation (BLC) and wide dynamic range (WDR) performance. An example is the Sony IMX-035 sensor, which yields noticeably different images with the BLC turned on in Image B and with the WDR switched on in Image C.
We consulted with manufacturers who participated in the test, as well as those who did not enter their cameras. The general consensus was that the ISP's software development kit was insufficiently open, resulting in wide-ranging image results.
Another factor may have been limited budget for development, or that the complementary encoder had insufficient image processing support or efficiency. Some product manufacturers felt that sensor suppliers were deliberately keeping the full capability of their sensors under wraps to maintain their competitive edge. A camera that was able to overcome the sensor's preset performance limits would attest to the R&D prowess of the manufacturer's R&D team.
There is no way to authoritatively confirm or deny these comments from manufacturers. Each one takes the same materials — similar to making an artistic masterpiece — but the results differ based on creativity. This is not to say certain cameras outperformed the rest, as each maker has unique design priorities. A variety of image sensors were deployed from Sony, OmniVision and Micron. The differences depend on the application,environment but perform superbly for another. We hope readers understand the performance differences between cameras for real-life usage, as each one is “best” in different conditions.