fish-length-opencv

by @mahaamesha

Contents

• About
• Usage
• Result

About

I want to measure length of fish. I use opencv to process the original image. The core of the result is based on image segmentation process.

• Segmentation Process
  The method that I use for segmentation is watershed algorithm. I also create some adaptation so that the result become more accurate (check segmentation.py).

• Measure Length Process
  To measure the fish length, I try to use 3nd degree polynomial fitting. This make good result enough for sample image.

  The other method I have tried is skeleton method, but the skeleton not coverage edge-to-edge of the fish.

  Fitting method can handle that, but the limitation is the function equation only give good result for landscape image.

• Validating Process
  To validate raw measurements, I use elimination of datum based on dynamic average value. This will remove small data whose value less than under_limit. To handle close objects, generally segmentation will fail, so for large data whose value greater than upper_limit, I divide them by it factor. Check validate_data(data) in mafunction.py.

  This will give more accurate result if there are N > 1 data.

  # This will be better if len(data) > 1
  def validate_data(data):
      ...
          elif n > upper:
              factor = int(round(n / avg, 0))
              copy.remove(n)
              for i in range(factor):
                  avg = np.average(copy)
                  copy.append(avg)
      ...

• Further Process
  I prepare the result to be saved as .json and .bin file. So, when I need to save the result to database, I can extract the measurement result from json file and the image from bin file.

  Bin file is encoded string from the image. I can save the encoded string to database. To show the image, I only need to decode the string.

Usage

To run this system, type this below in CMD.

• For v1.0.0:

  python main.py -i img/lele.jpg -w 1
  

  I set the default -i for lele.jpg with -w 1. So, I also run it by python main.py sometimes.

• For v2.0.0:

  python run.py
  

Result

This is the result.

Reference

• Image Segmentation with Watershed Algorithm
• IMAGE SEGMENTATION AND MATHEMATICAL MORPHOLOGY