Image Processing Operations

[parth-p]

System information

• TensorFlow version: 2.0
• TensorFlow Addons version: 0.3
• Is it in the tf.contrib: No
• Are you willing to contribute it: Yes
• Are you willing to maintain it going forward?: Yes

Describe the feature and the current behavior/state.
I am proposing the following image operations:

1. SmoothingFilters: https://colab.research.google.com/drive/1SnSQ-mOsglB9nHNFp1YkCgEk4iR5qa7n (Implemented Averaging filter and Gaussian filter for smoothing images).
2. Thresholding: https://colab.research.google.com/drive/1CdVfa2NlkQBga1E9dBwHved36Tk7Bg61 (Implemented basic thresholding, adaptive thresholding, and Otsu's thresholding method).
3. GradientFilters: https://colab.research.google.com/drive/1MpKgEIYnXPjdKGQQ9sVzcMna33_UCO5T (Implemented Laplacian Filter and Laplacian of Gaussian filter).
4. Canny Edge Detection: https://colab.research.google.com/drive/1CadYuEWvSORYDyRKxBhCPKWf7c1w04wf (Implemented Canny edge detection method)

[WindQAQ]

Hi Parth,

1. Averaging filtering has already existed in addons.
2. Other operations look good but I do want to confirm if addons should take them all. @seanpmorgan Sean, how do you feel about these operations?

Thank you!

[parth-p]

@WindQAQ, Thanks for notifying me on this one, I'll remove the Averaging filter part from the notebook.

[seanpmorgan]

Hi Parth, do you have any information on some of these being using in a ML pipeline? I'm afraid it's hard to justify all standard image operations becoming part of Addons.

I think we could work in most of these if we had a solid use case (e.g. image preprocessing for pathology). Perhaps if you could update these examples to show how they can be used in a regression problem... how many cells per image or something

[seanpmorgan]

@parth-p Could you go ahead and submit a PR for thresholding methods?

[parth-p]

In addition to the previous image operations, I have implemented a few more operations:

1. Hough transform Lines: https://colab.research.google.com/drive/1PCvihgwulfcNeDGmqrqAA9pN9oNiWJYd (Detecting straight lines in a given Image)
2. Hough transform Circles: https://colab.research.google.com/drive/1q_Q6isn7l7fgoqWA6FgNw6OegCc0TjbM (Detecting circles in a given Image)

Also, there is an update in the implementation of canny edge detector.
Link : https://colab.research.google.com/drive/1Emd7T5tOMUjapcwCReZlKKWUpA8Yw__b

[bhack]

We are making a catalog of some useful image ops for AutoAugmentation policies at #1333. Probably some ops could be useful also in post-processing tasks as i.e. Mediapipe heavly use Opencv operation in postprocessing calculators. See google-ai-edge/mediapipe#515

[Squadrick]

Closing this issue due to lack of activity. Discuss on #1333, that seems to have grown into the mega-thread for all things image proc.