Re: [GSoC] plans ahead

From: Miral Shah (miralshah2211_at_[hidden])
Date: 2019-07-27 19:15:29

On Sat, Jul 27, 2019 at 12:19 AM Mateusz Loskot <mateusz_at_[hidden]> wrote:

> On Fri, 26 Jul 2019 at 17:25, Miral Shah <miralshah2211_at_[hidden]> wrote:
>
> Miral, yes I suggested you to sort your TODO list according to decreasing
> complexity and pick the first item (most complex) first.
>

The remaining task in the descending order of the complexity:

   1. Implementation of 2D convolutions
   2. Gaussian Adaptive Threshold
   3. Image denoising (Predefined kernels can be helpful in this case such
   as median filter, average filter or Wiener filter)

Please, give an overview of what you are planning to implement as
> part of the 2D convolution and kernels.
> GIL already provides related features. Olzhas has contributed some
> more. It would be good to list exactly what is planned to be added.
>

In this, I would like to maintain APIs similar to the existing 1D kernel
and the convolution.
To achieve the goal new APIs for 2D is as follows corresponding to their 1D
counterpart:

   1. kernle_1d --> kernel_2d
   2. kernel_1d_fixed --> kernel_2d_fixed
   3. convolve_row/convolve_col --> convolve_2d

As a byproduct of the convolution correlation function can also be
implemented this really depends upon how we chose to implement the
convolution. Current numeric implementation of convolve functions is
implemented using the correlate_row/col.

> I'd suggest to open a new issue  "[gsoc2019] 2D convolution and kernels"
> and outline your plan there.
>

 sure.

-- 
Thank you,
Miral Shah