$3$ Implement a Demosaicing Algorithm

Implement the simple demosaicing algorithm seen during the lecture, or a more advanced one.

Apply it to IMG $4782.$

Important: in the processing pipeline, do not change the data type $($i$.$e$.,$ do not transform

it to unsigned $8-$bits integers$).$ Conversion to a less accurate type should be done only when

exporting results to an image format. Hint: no need to apply a median filter on U and V

channels; it is slow and won$$t improve much the image. Hint: if it looks grayish, it$$s normal:

you haven$$t applied some color balance yet.

$4$ Improve the Luminosity

As you probably noticed, the image you obtained after demosaicing is quite dark. Apply a

gamma correction $$ I suggest $\backslash$gamma $=0\mathrm{.}3,$ and not to normalize using np$.$min and np$.$max but

np$.$percentile with values of $0\mathrm{.}01$ and $99\mathrm{.}99.$

Evaluate at least one other curve than y $=$ x$\backslash$gamma $.$

Hint: use the percentiles to normalize data in $[0,1]$ as you would normally do with min and

max $($set negative values to $0,$ and values greater than $1$ to $1),$ apply the gamma correction,

and invert the transform to get back the the previous range of values.

Hint: here is how to use the percentile approach:

a $=$ np$.$percentile$($data$,0\mathrm{.}01)$

b $=$ np$.$percentile$($data$,99\mathrm{.}99)$

data $=($data $-$ a$)/($b$-$a$)$

data$[$data$<0]=0$

data$[$data$>1]=1$

$5$ White Balance

The last step for processing raw sensor data is to apply white balance. Implement the gray

world in your algorithm.

Hint: because of the multiplications, some values might become greater than $255$; clip them

to this maximum.

$6$ Initial HDR Implementation

This exercise is the initial implementation of HDR images, as seen during the lecture. You

will find on StudOn a set of photos $($from $00.$CR$3$ to $10.$CR$3).$ For each of them, the exposure

time was half the one of the previous photo.

Combine them to produce HDR raw data. Then, apply the demosaicing algorithm and the

white balance. Decrease the dynamic range by computing the logarithm of this data, and

downscale it to the $[0,255]$ interval. Then save the resulting image.

You should obtain the same result as what is shown in the slides.

$7$ iCAM$06$

Implement the iCAM$06$ HDR method, as explained on the slides. Try different settings to find

the ones you like the best.

Hint: bilateral filters are slow, so while developing your solution, use at first very small kernels

to minimize the run time.

$8$ Win

$..$ or at least try to$!$

Implement a demosaicing function named process raw which takes two parameters:

$1.$ The path to a raw file $($CR$3$ format$)$

$2.$ The path to wich the resulting JPG image has to be saved

Hint: you will be judged based on the quality of your image, so when saving it to the JPG

format, set a high quality setting, such as $98$ or $99,$ not $75!$

Provide code for each one differentely $3,4,5,6,7,8$