PHOTOSHOP - A couple of simple additions to the Custom filter for completeness, competitiveness

  • 1
  • Question
  • Updated 4 years ago
  • (Edited)
The Custom filter offers only half of what every other convolution matrix filter offers; it needs a factor and a bias input (offset makes no sense; there's no equivalent to that in matrix calculations performed on images—is there?).

Also, many of the commonly used edge detectors are applied in two directions (horizontal and vertical); and, other edge detectors are only useful with gradient thresholds are applied to them. By adding these two abilities to this filter—plus all of the inputs requisite for a complete convolution matrix solution—would not only bring this filter up to par with others, but would place it far ahead.

Although I doubt many complaints are made about this filter, I do know that it's not widely used (some would say it's too complicated or doesn't make sense); but, for those who can use it, this request describes the features necessary to use it properly, effectively and powerfully.

Here's a link that describes one use of a convolution matrix used in image processing; as you can see by the box blur, Gaussian blur and Unsharp (mask) convolution matrices, there is more to a matrix than just the rows and columns of numbers:

The fractions outside and to the left of the matrix are factors. The other feature requests described here are explained in ImageMajik's feature description page (for example):

As you can see, there are a number of operations that are applied in conjunction with convolution matrix operations, and occasions when two matrices are applied consecutively. Note that ImageMajik goes beyond offering these by categorizing uses of matrix operations, overcomplicating the filter, and discouraging new users to make an effort to understand how convolution matrices work. To them, it looks as if it already offers a complete solution, when, in fact, it limits users to only the ones it offers. A complete and competitive Photoshop filter-equivalent does not need the additive complexity by matching the ImageMajik feature-to-feature; as I said, a couple of extra inputs offers everything. The users who would use the Custom filter would already be familiar with all the ways matrices can be used, and only need basic inputs.
Photo of James Bush

James Bush

  • 1 Post
  • 0 Reply Likes

Posted 4 years ago

  • 1
Photo of Chris Cox

Chris Cox

  • 20280 Posts
  • 823 Reply Likes
Offset == bias
scale == factor
Yes, those are standard things used in convolution kernels.

All of the kernels given on that wikipedia page can already be implemented in Photoshop's Custom filter.