Aurora 2.1 - Photoshop Plugin Setup Free
Shanae Maerz
This is a guide on using Aurora plugins (www.aurora-plugins.com) for performing the measurments and for computing the two FIR filters which perfectly equalize the response of your headphones ON YOUR OWN HEAD. This is a "personal eq", which also removes the "pinna effect" of your ears. The resulting equalization is only suitable for listening to binaural sound, not to normal stereo.
In fact in any listening experience only ONE SET of pinnae should be present. In normal stereo listening, these are YOUR pinnae, and you should not remove their effect. Instead, when listening to binaural, the HRTF is already incuded in the sound being reproduced, so the colouring effect of your pinnae is detrimental, and should be removed.
2) the binaural recording is a rendering of anAmbisonics orSPS soundtrack, done employing your own individualised HRTF set, coming from measurements in an anechoic room or from computer simulations based on 3D scanning of your head and pinnae.
However, even when listening to binaural recordings made employing a "generic" HRTF set (for example the widely employed "dummy head"Neumann KU-100), your pinnae are detrimental to the listening experience, hence employing the FIR eq filters, computed following the procedure described here, becomes highly beneficial, both in terms of spectral neutrality of your headphones and in terms of spatial accuracy of the sound sources.
The hardware required are your own circumaural headphones, a pair of endoauricaular binaural microphones (possibly a DIY set) and a decent audio interface featuring good headphones output and stereo mic inputs (providing power to the microphones, if required). The following photos show the hardware required:
Pressing OK the sine sweep is generated, but, as shown here, the Inverse Sweep is also generated and automaticlaly loaded onto the Windows clipboard (be aware that Audition can also use a number of "internal" clipboards, hence if you want to access the Windows clipboard, for pasting the Inverse Sweep to a new waveform, you have to set Audition for using it, with the command Edit - Set Current Clipboard).
We can now deconvove the impulse response, by convolving the recorded sweep with the Inverse Sweep, which is still in the WIndows clipboard. The convolution is performed invoking the Aurora Convolve with Clipboard plugin:
As shown, the Full Autorange is selected. The result of the convolution is a very long audio track, but we can select and trim just the portion around the main linear peak, as shown here, where a portion of just 1024 samples was selected:
For checking how well these inverse filters behave, we come back to the measured impulse responses of our headphiones + ears, and we apply the Convolve with Clipboard plugin, so that we equalize them with the inverse filters.
The case 1) has already been described at the end of previous chapter, when we convolved the measured IRs with the inverse filters. Now instead of the measured IRs, you simply open your binaural recording, and convolve it with the clipboard (where of course you have copied the EQ filters). Only trick is to specify that the convolution has to be done selecting both channels of Audio Data and of Impulse Responses (filters), as shown in the screenshot in the previous chapter. DO NOT SELECT the Crosstalk Cancel Mode, this is for loudspesker reproduction, here we are addressing headphones, which have virtually no crosstalk.
So we present here case 2). We assume one has a good HOA recording. In this case it is a third order Ambix soundtrcak (16 channels) obtained recording an opera on-stage using anEigenmike-32(TM). We want to listen to it with our headphones, and using anMbientlab MMR headtracker, driving the IEM Scene Rotator plugin employing Hector Centeno's OSC bridge namedMMR2OSC.
The only trick is to specify that the filter set is "diagonal", meaning that each of the two inputs to X-MCFX has to be convolved with just one filter and sent to the corresponding output.
This is done when selecting the filter matrix to be employed in X-MCFX. One first clicks on the button with three dots and navigates to the folder containing the filter(s).
Then clicking on the Filter drop-down menu, one selects the proper WAV file containing the filter matrix. At this point the following pop-up windows appear, asking for the number of input channels or for the "diagonal" option. Of course we select the second one:
The CPU load caused by the convolution of these two short filters is truly minimal, so this approach can be used even on old computers with limited capabaliities, thanks to the super-efficient convolution algorithm employed in X-MCFX.
A total NOOB to photography here. I was in Troms, Norway this weekend and was lucky enough to capture some pictures of the lights. Now, when I say I'm a noob, I actually mean that I know very, VERY little about photography and mainly use my Canon 70D to shoot up-close videos. I set my aperture to the lowest it would go (3.5 with the kit lens) and the shutter speed to 10 seconds, ISO to 1250. I captured some dramatic lights but unfortunately these two are just out of focus. Is there any way to sharpen them without too much distortion / noise? I don't own photoshop or any other photoediting tool really. What I mainly am looking to do is just have the stars appear sharper/clearer.
This is an easy task using deconvolution, as the stars in the image should be sharp points but they are blurred. This means that the stars in the image can be used as the point spread function that you need to perform deconvolution with. You can do this using the free of charge program ImageJ and the plugin parallel iterative deconvolution or DeconvolutionLab. The image must first be converted to linear colorspace and split into the color channels. You then perform the deconvolution on the color channels separately and then recompose the color image. And then you convert the deconvolved image back to sRGB colorspace.
I've attempted to sharpen the picture using deconvolution, but to get really good results requires a lot more work and also one has to use the raw file because the jpeg file has degraded information about the high frequency components. There are three major problems one faces in this particular case. The first is posterization effects (visible discrete jumps in brightness) which are also due to working with the 8 bit jpeg instead of the 16 bit tiff file (which is actually just obtained from the usually 14 bit raw files).
Another problem is caused by the sharpening of the stars causing them to exceed the maximum brightness for 8 bit images. While the stars in the original image were not overexposed, they will become de-facto overexposed if you use the standard method of clipping brightness values higher than 255. If you just led the brightness be whatever it becomes, the entire picture will become way too dark. Now, clipping the brightness values at 255 leads to two problems, one is that you lose the color of the stars.
Clipping also leads to another problem. The stars which after deconvolution had a sharp brightness profile will now appear to have a flat brightness profile, not all that different from how they appeared in the original image. Note that the whole point of this exercise as requested by the OP was to sharpen the stars!
The first problem can be dealt with by transforming the image from sRGB to 32 but linear RGB (so, we make sure not to transform to 8 bit linear RGB) and then when transforming back this is better done using a custom made gamma correction instead of the standard transform.
The problems caused by the brightness of the stars after deconvolution can be dealt with using tone mapping methods similar to what one does when processing HDR images to 8 bit images. This doesn't fully correct for the problems, the brighter the star the less sharp and the more white it will appear. Also deconvolution artifacts will become visible.
Zoom into the image and select a bright star somewhere in the middle of the image, duplicate the image (which in ImageJ will create a new image with just the selection). Then measure the background in the color channels by selecting a part of the background and using the "measure" function. If the background is reasonably uniform you can subtract this value from the gray values in some color channel and you have a reasonably accurate point spread function. To do this really accurately, you should repeat this for a few stars nearby, align the images and take the average.
Also, due to part of the unsharpness coming from the long exposure time causing the stars to become star trails, the point spread function is not uniform throughout the image. This requires the use of many different point spread functions for different parts of the image. The mountains in the distance should be sharpened using a different point spread function based on the width of the blurry stars.
I just compiled my point spread function based on the image of just one star, so a there is a lot of room for improvement here. Then the next step is to split the 32 bit RGB image into its color components, by selecting "images to stack" and then you run "parallel iterative deconvolution", I selected the MRSND method as this gives the best results for dim images (leads to less artifacts which combined with posterization can look really ugly), for preconditioner select "none", the boundary can be taken to be zero as the boundaries are quite dark, this won't lead to artifacts (otherwise selecting reflexive will usually be better). The we want to do the calculations as accurately as possible, so it's best to select "next power of two" for resizing, output "float" but in this case this is the same as selecting "same as source". Precision should be set to double. For MRSND the maximum number of iterations should be quite large, I've taken this to be 50. The maximum number of threads depends on your computer, it's usually preset to the right value for your computer.
Then you click on "deconvolve", things may then go wrong due to an out of memory error. In that case you can reduce the precision from double to single, or you can cut the image in parts, deconvolve the parts separately and at the end stitch everything together. 8 GB RAM is sufficient for this image, but for slightly larger images you can bump into this problem with only 8 GB RAM.
05f2edc126