Monday, 13 May 2013

Review...Canon Eos M High ISO Noise - Part Three


Canon Eos M 18mp Aps-C size sensor, how does it perform at high iso, noise and some suggestions on how to eliminate it . . .

Read Part Three of my review.



ISO 3200 









In essence high ISO files are just underexposed images which are then corrected in software. It really just means that the dark image is made lighter via the camera software so that it looks correctly exposed. Just like in the days of film,  underexposed shadow areas look gray and grainy. The better the sensor and software the better these underexposed images look like. Software plays a big part and can best be illustrated by the fact that the first Sony 24 megapixel sensor as used in the Sony Alfa 900 and  the Nikon D3X, a camera generation ago, but where Nikon managed to get vastly superior high iso files than Sony could.


The larger the sensor the less noise it produces and the smaller the sensor the more noise it has. Thus a full frame sensor like that of the Eos 5D mk2 has less noise than the APS-C sensor of the Eos M. As more and more  pixels are crammed into the sensor it has a bearing on the noise levels and thus, pratically, an 18mp Aps-C sensor has inherently more noise than that of the same size sensor but with fewer pixels or a full frame sensor with the same number of pixels.

Every new camera that's announced has more pixels than the one it replaces. Pixel counts are market driven because we the consumer look at pixels like we look at the cubic inches of a car engine, the more must be the better...

One exception is the Canon Eos line of camera which has had the same sensor for the last few years with only minor tweaks for each new successive model. Canon is known for their 'noisy' sensors, especially shadow noise at higher iso's and also the 'banding' found in some sensors as compared to the new sensors from Sony and Fuji...but that doesnt mean that the sensors from Canon are useless, it just means we have to find ways of working around that...and fortunately there are a couple of noise reduction programs that can help.

As mentioned before this is my first high megapixel Aps-C sized sensor. My Canon Eos 20D and 40D 'only' had 8 and 10 megapixels respectively and were know for their 'clean' sensor design. My Leica X1 'only' has a 12 megapixel sensor, apparently identical to  the Nikon D300.

....So my personal rule of thumb is that 'more pixels add more resolution but also more noise at higher ISO's'...may would disagree with this statement(and some have indeed disagreed and with scientific papers to back up their claims) but it is 'my' rule of thumb, and all sensors being equal ito current technology and pixels 'I' would opt for the lowest number of pixels to give me less noise at higher iso's... 


First Test

For my first test I chose a dark corner of my office and photographed my gardener, Exodus standing against a grey filing cabinet. Exodus is from Malawi and the further north one goes the darker the skins.

Lighting was a mix of daylight, fluorescent and incandescent. I used the white light switch for neutral balance. I used Lightroom 4 to process the raw files and used my standard settings for sharpnening, of Sharpness 50, Radius 0,5 and Detail 11.

At 800 iso noise is really not intrusive and I would use it without any noise correction. At 1600 iso noise starts to creep in and at 3200 iso noise is quite visible but at 6400 and 12800 it becomes, imo, quite unacceptably noisy. I would use 1600iso at a push but 3200 is already too noisy for my own preferences. I regularly use my  Eos 5d2 at 1600 and 3200iso...With the Eos M I would be more reluctant....

Below is the test images. On left the image before before noise reduction and on right the same image  after applying noise reduction. I used a very useful plugin from Topaz called Denoise. It has standard settings like Raw -lightest, -light , -moderate etc which can also be tweaked to taste. Before opening the images up in 'Denoise' I reduced the sharpness settings to 'Zero' in LR. After 'de-noising' I dialed in a small amount of sharpening in LR4.

All the images posted here we run through the Topaz Denoise plugin as described above. Generally I used the 'lightest' noise reduction setting. If some luminance noise is still visible its because if I reduced it more, the sharpness would be affected. Noise reduction setting are very much a matter of taste and each one will probably find their own balance between acceptable luminace noise and sharpness. Chroma noise is much easier to correct and I just left the LR4 as is before running it through Topaz Denoise.

Luminance noise is the big culprit here and it looks a bit like 'grain'. The higher the ISO the more luminance noise. Interestingly the detail in the fabric of the shirt still looks fine even after the noise reduction...it is the face that starts to loose detail and texture and at higher ISO's, after noise reduction, starts to look mushy' and with a blurring away of details.

800 iso
                                       
800 iso cropped to same size as images below.
   
ISO 1600






3200 ISO




6400 ISO



12800 ISO





Another test.



The image below is typical of the type of image that I would do indoors handheld. For the image to be sharp I would need a shutter speed of at least 1/30sec, in this case 1/60sec. Aperture was set here at  f3,5 which will ensure a little more depth of field than wide open so that the eyes and a bit more will at least be in focus. To achieve these setting I needed an ISO of 1600.


At 1600 ISO the eyebrows are also relatively sharp and at 100% enlargement one can just see individual hairs. At higher ISO's the detail starts to blur, so it looks like 1600iso is about the max for close up detail images like this one. The more expansive the subject  matter the lower the ISO would have to be to still capture fine detail.


In Post Processing I still 'pushed' the exposure by another 2/3 stop to see how the shadow areas would hold up. They look fine to me. To retain highlight details I pulled over the 'highlight' slider to '-100', which  makes the image slightly 'flat' looking.


I then reduced the noise via topaz Denoise and chose the 'Raw Light' preset which is one up from 'Raw Lightest'.


Back in LR, I increased the 'clarity' by +50 to bring some 'punch' back into the image and sharpened the image a little bit for the first time. This is typically how I would normally process an image like this...





1600 iso


1600 iso.





'Real world' Test

Ok one can just do so many boring test images...so here are couple of 'real world' samples...btw somehow 'real world' has become a byword for something that is inherently of greater value and closer to the truth....I am not sure if I can agree with that but here goes....

Barbora Tellinger a well know jazz singer and lecturer her in my part of the world asked me to take some photographs of her last concert at the Musaion music center of the University of Pretoria. I was asked not to leave my seat as they were also making a video of the proceedings. I sat a couple of rows from the front in an isle seat so that I could set up my tripod. To get close up images I needed my Canon 70-300L zoom and attached it to the Eos M via the lens adapter. I shot at a variety of Iso's up to 6400. Initially I had set the camera to 'Scene Intelligent Auto' a sort of super 'program' function. The pics came out fine with good exposure of the white skinned singer against the dark background. Afterwards I changed the setting to Aperture priority AE so that I could ahve ore control over the aperture and ISO. In hindsight the 'Scene Intelligent Auto' worked best and it just goes to show that these cameras are made to take the 'thinking' away from the operator...and leave it all to the camera. and actually it works very well and under many conditions the camera does just fine. This is the way it was designed to work best and its when we want to impose our will on the camera where things become more difficult and cumbersome. 

So the moral of the Eos M? Just leave it to the camera to do your thinking....it will probably lead to good if not better image quality than one can do by trying to control the camera too much...

What these images have clearly shown me is that the practical limits of high ISO is determined by the subject matter and the size of the object being photographed...as you will see ISO6400 may be ok for a closeup of a singer on stage but its definitely not ok for 'landscape type' images...



ISO 5000 f5.6 1/400sec after noise reduction and detail below..









ISO 1600 detail and original below



ISO 3200 detail and original full frame file below. 

I made a beautiful detailed  610mm x 430mm print of this image.
The file was good enough to handle extensive post processing...
Detail ISO 1600, quality starting to drop, original below. Moire in shirts



Moving away from the jazz concert to my local shopping centre...I found that images like this one below shows up the limitations of high iso shooting. There is just too much detail in the image for the sensor to be able to render them sharply at high iso. These images were made at 3200 ISO wide open on the 22mm f2. Its just not good enough.....and anymore luminance noise reduction just blurs the detail even more


Detail, ISO 3200 at f2 with full frame image below.

     



Conclusion.

Why use high iso settings? To get a high enough shutter speed to freeze motion or prevent camera shake. If you don't need a high shutter speed its best to put the camera on a tripod and use the lowest iso possible...the lower the iso the better the quality.


These test have clearly shown that iso 6400 may be good enough under the right lighting conditions and fairly close up,  but that even 1600 iso may be too much for certain adverse conditions. What those conditions are depends on one's own personal idea of quality and whats acceptable or not....


Also it depends what size the final image will be viewed at and on what medium. Canvas will probably be the best for large prints  and when the paper quality gets better as with satin or glossy media the prints will have to be smaller to hide the noise... 


Regards,


Ivan