The transition form CCD to CMOS technology is a convenient demarcation in the evolution of digital imaging. But this doesn't mean perceived rendering differences are directly related to technical differences between CCD and CMOS photo-diode arrays.
As
Benjamin Marks mentioned color rendering is a very complicated issue. While perception preferences are authentic, it is difficult to make definitive conclusions about the basis for the differences.
There is no need to invoke CCD technology to account for personal preferences in color rendering. Color rendering differences between cameras can be caused by many factors. The CCD and CMOS design architectures and manufacturing processes are very different. But these differences have a negligible impact on color rendering compared to other sensor assembly hardware and demosaicking algorithm differences.
Both CCD and CMOS photo-diode arrays produce the exact same thing - photoelectrons. All photoelectrons are identical and they do not contain any inherent information about color.
The spectral response of CCD and CMOS photo-diode arrays are not identical. However the differences are small. It turns out typical CCD arrays are more sensitive to near IR and are slightly less sensitive to blue wavelengths. These differences are cause by physical fabrication differences before the light reaches the photo-diode.
However, the Nikon D100, D200, D40 and many other early digital cameras have CCD sensors. So superior perceived color rendering due to CMOS-CCD spectral response differences should be ubiquitous. This doesn't seem to be the case.
Other factors such as the transmission properties of the color-filter array assembly and the IR filter transmission properties can affect the relative number of photoelectrons generated R, G and B channels.
Here is the CFA spectral response for two Canon DSLR's. There is considerable overlap between the R, B and G filters. This produces contamination (cross-talk) between the three channels. When the CFA frequency response is narrower, less light reaches the photo diodes. But there is also less contamination. The opposite holds as well. Brands may have decided signal-to-noise was a more powerful marketing message than color rendition. So, older cameras (CCD) may have superior color rendering compared to newer cameras (CMOS) simply because the older CFA designs were optimized to maximize color rendering at the expense of SNR. Another important factor is smaller pixel pitches exacerbate the spectral cross-talk between photo sites. By coincidence CMOS cameras have smaller pixel pitches because resolution became a marketing priority as CMOS technologies matured.
The IR filter layer, the higher NIR spectral response of CCD photo diodes and the CFA red filter spectral response all impact the relative number of photoelectrons for the red channel. A strong IR filter could negate any increase in red signal levels due to CCD NIR response. If the spectral response of the red filters in the CFA was asymmetrical to include lower frequencies (NIR) the red channel signal levels could be relatively higher.
But wait, there's more.
During image rendering the camera color profile has a strong impact on perceived color rendering.
The Bayer demosaicking algorithm mathematical model assumes there is no color contamination between the R, G and B channels. Camera color profiles are used to model the unavoidable cross talk so the total demosaicking model matches the data. These correction models are proprietary. The practical aspects involved in optimizing a demosaicking algorithm mathematical model are complicated.
Here is a nice review. As CFA cross-talk increases, the impact of the cross-talk correction models increases (
link). The differences in perceived color rending between cameras will depend on color profile mathematical models.
