As mentioned earlier, the Epic Dragon offers a wide range of different compression ratios, which allow various frame rates. We chose to analyze the camera’s output at a low 5:1 compression ratio, allowing for a high frame rate while also preserving a high level of detail.

Spatial noise reduction?

As illustrated in the following graphs, we did noticed a very slight spatial correlation in some parts of the dynamics, but nothing heavy-handed enough to impact score significantly. This is a real plus for image quality, as noise reduction filters lower spatial resolution and degrade the effectiveness of a RAW convertor’s noise reduction capabilities (such as DxO Optics Pro’s Prime Noise Reduction).

IMG_19022014_190225

The example above shows autocorrelation for gray level at 18% of the sensor’s dynamic response on the green channel. The single peak clearly indicates that no spatial noise reduction filter is being applied in that case.

IMG_19022014_190235

A correlation between the pixel in the center and the 4 neighboring pixels, in this second sample, indicates the possibility of very slight smoothing at a lower percentage of the dynamic. This was taken into account to evaluate RAW noise before spatial noise reduction

The highest color sensitivity ever measured: exceptionally low noise at the base ISO.

IMG_19022014_190244

SNR comparison between the Epic Dragon and the Nikon D800, both at their base ISO (Shown in print mode, normalized to 8Mpix) SNR (in dB) as function of the dynamic (in %)

The headline news is, of course, that of the Dragon’s exceptionally low noise levels. The SNR (signal/noise ratio) curve shown above is simply outstanding - it’s higher than the Nikon D4’s curve for the whole of that sensor’s signal response. The shape of the response curve is interesting as well.

Close to saturation point, it shows a kind of plateau or leveling off, which indicates that the SNR is mostly limited by some pixel response non uniformity and not by the photonic noise.

This noise curve, coupled with a good color response, leads to a very high color sensitivity. At the pixel level (screen mode), the color sensitivity is simply the best ever measured. We normally calculate color sensitivity using an 8-bit output sRGB colorspace. In this case, however, the 8-bit sRGB color space was not large enough to accurately represent the color sensitivity of the Epic Dragon. As a result, we had to switch to a 16-bit sRGB color space!

Spectral response: high sensitivity of the RED channel.

The color response of the sensor shows an unusually high sensitivity on the red channel (see graph below). This certainly contributes to the overall sensitivity of the sensor.

Also, in daylight illumination, red and green are saturating for approximately the same exposure. This yields a very low amplification of the red channel, with respect to the green that will be needed for white-balancing (In D50 illuminant, the red white balance scale is only 11% above the green one).

L_E19.tmp

Red Epic Dragon sensor color response. The 3 channels are normalized with the highest sensitivity measured

Sensor size and multiple sampling.

With a sensor area 0.56 x smaller by surface area than a 35mm (24 x 36 mm) full frame sensor, the results are very impressive. As the sensor and image processor can deliver very high frame rates, the Epic Dragon is certainly adopting multiple sampling techniques to reduce noise levels (also known as temporal noise reduction). Without such processing, such high SNR would only be possible from a sensor with an exceptional Full Well Capacity. Performances like this seem, to us, above the current technical capabilities of CMOS sensors.  

As a side note, it’s interesting to speculate whether rivals such as Nikon, Canon or Sony are already adopting such techniques during video capture or in jpeg. But, this is the first time we can assess this type of performance on still RAW.

 

Digital Gain and dark uniformity in the field: slight weakness.

Contrary to the more conventional still camera which adopts analog gain, cinema cameras like the Epic Dragon do not have to provide such features as most footage is captured in good lighting (or controlled) conditions. The Epic Dragon was designed with this purpose in mind.

It’s also worth noting that the Epic Dragon only has one ‘native’ ISO setting (ISO 250 which is measured at 104), the others are achieved by applying digital gain.

To realize the full DxOMark sensor analysis we had to apply digital gains and then stress the darker part of the dynamic.

And so the problems begin….

First of all, a very small row/column noise was detected on the dynamic darker part, as you can check with the line detected on the autocorrelation shape below. This very small row/column noise does not impact the incredible dynamic range at the base ISO, but as digital gain is applied to achieve higher ISO settings, it becomes significant, affecting noise levels (and the respective amount of spatial correlations), thereby decreasing dynamic range (as you would expect).

IMG_19022014_190405

We also noted a dark non-uniformity in the field: the mean value can vary from 1 to 3 grayscale shades (for a 16-bit per sampled pixel). While this isn’t important for base ISO, we had to take it into account extrapolating the higher ISO sensitivity settings.

Conventional DSLRs benefit from Analog gain that seems to amplify read noise and fix pattern noise in a smaller way than a pure digital Gain. This explains, despite its quality, why the Epic Dragon has lower low-light performances than the top DSLRs. So for cameras, where the lighting can’t be controlled, analog gain is preferred. That’s not such a high priority with a cinema camera, but the Epic Dragon produces very clean looking files and can be used safely up to nearby ISO 3000, with an excellent image quality.