Added Hough Line Transform #601
Conversation
src/houghtranform.jl
Outdated
| min_theta < max_theta || error("max_theta must be greater than min_theta") | ||
|
|
||
| if ! is_binary_image(img) | ||
| img = canny(img) |
There was a problem hiding this comment.
This breaks orthogonality; what if someone wants a different algorithm than canny? Will be solved if we simply require a Bool eltype.
Hmm, I hadn't noticed that canny returns an Array{Float64}. We should probably change it to return an Array{Bool}.
src/houghtranform.jl
Outdated
| rho > 0 || error("radius threshold must be positive") | ||
| min_theta < max_theta || error("max_theta must be greater than min_theta") | ||
|
|
||
| if ! is_binary_image(img) |
There was a problem hiding this comment.
Better to just require an T<:Union{Bool,Gray{Bool}} eltype.
src/houghtranform.jl
Outdated
| img = canny(img) | ||
| end | ||
|
|
||
| width::Int64 = size(img)[2] |
There was a problem hiding this comment.
No need to declare types like this.
src/houghtranform.jl
Outdated
|
|
||
| function hough_transform_standard{T}( | ||
| img::AbstractArray{T,2}, | ||
| rho::Float64, theta::Float64, |
There was a problem hiding this comment.
Too restrictive on the types? Particularly for people who have 32-bit machines, Int64 is not native (better to use Int). Number and Integer might even be best.
src/houghtranform.jl
Outdated
| width::Int64 = size(img)[2] | ||
| height::Int64 = size(img)[1] | ||
| irho::Float64 = 1 / rho; | ||
| numangle::Int64 = round((max_theta - min_theta)/theta) |
There was a problem hiding this comment.
round(Int, x) returns an Int. You can use this in several places below, too.
src/houghtranform.jl
Outdated
| fill!(accumulator_matrix, 0) | ||
|
|
||
| #Pre-Computed sines and cosines in tables | ||
| tabsin = Array{Float64}(numangle) |
There was a problem hiding this comment.
Even better, Vector{Float64} (here and below).
src/houghtranform.jl
Outdated
|
|
||
| #Hough Transform implementation | ||
| for pix in CartesianRange(size(img)) | ||
| if img[pix] == 1.0 |
There was a problem hiding this comment.
If you're using Bool eltypes, this becomes just if img[pix]
src/houghtranform.jl
Outdated
| if img[pix] == 1.0 | ||
| for i in 1:numangle | ||
| dist::Int64 = round((pix[1] * tabsin[i] + pix[2] * tabcos[i])) | ||
| dist += round((numrho -1)/2) |
There was a problem hiding this comment.
This constant could be hoisted out of the loop.
src/houghtranform.jl
Outdated
|
|
||
| #Finding local maximum lines | ||
| validLines = Array{CartesianIndex}(0) | ||
| for val in CartesianRange(size(accumulator_matrix)) |
There was a problem hiding this comment.
Perhaps we should add a function input to findlocalmaxima and just use that?
src/houghtranform.jl
Outdated
| #Sorting by value in accumulator_matrix | ||
| sort!(validLines, by = (x)->accumulator_matrix[x], rev = true) | ||
|
|
||
| linesMax = min(linesMax, size(validLines)[1]) |
There was a problem hiding this comment.
can just use length instead of size(validLines)[1] (they're equivalent if validLines is a vector).
|
Is there any function available for drawing lines using the normal parameters generated by hough transform? I found the package Cairo.jl but is there something simpler than that. There is an ImageDraw.jl package and a line function implemented in it. Should I write a function to draw line from normal parameters which uses this existing function? |
|
@timholy Yes, I also think it would be better to have it in ImageFeatures.jl instead of having it directly in Images. Should I create a new pull request in that repo? Also what updates are required in the ImageFeatures.jl package related to the new images framework, maybe I can help :) |
|
@annimesh2809, I suspect ImageDraw would indeed be the better way to go...fewer dependencies. Great, I think ImageFeatures is the best place. With regards to updating ImageFeatures, if you install it and type |
|
@annimesh2809 It would be great if you work on updating ImageFeatures to the new Images.jl. I have been wanting to do that for a long time but haven't been able to get time. I'd be happy to help in case you need any 😄 |
|
@mronian I would be really happy to update ImageFeatures.jl. But I did not understand what you meant by update ImgaeFeatures to the new Images.jl |
|
The deprecation warnings and test errors should basically tell you what to do. |
|
I have started working on updating ImageFeatures.jl to work with the new images. But there are some failing test cases without changing any code:
Should I comment out those test cases and create a log of them so that we can correct those test cases later when all the deprecated warnings have been removed? |
|
I guess some changes to Images.jl have modified some values of the descriptor. This should be solved if take an approximate equals instead of equals. Lets close this thread and move it to ImageFeatures.jl. |
|
@annimesh2809 Lets continue this discussion at JuliaImages/ImageFeatures.jl#24 |
|
Whoops. I did not realise this was a PR for hough lines. I thought it was just an issue. I'll reopen it. |
|
@mronian Actually we should close this PR because hough lines will be added to ImageFeatures.jl as discussed above once we make it function without errors. |
Added a new feature detection : Hough line transform as mentioned in #600 .
The method returns an array of tuples containing (r,theta) as parametric values of lines present in the image.