image-flow

About

Python image modification class using builder/method chaining design pattern. Originally intended for image preprocessing for ML models. Incorperates a suite of image processing operations from numpy, skimage, scipy image, and PIL.

Advantanges of image-flow

• Easy to perform multiple consecutive operations on an image
• Easy to visualize image (with im/imnz/rgb functions) in iPython for pipeline development

Special use-cases:

• Find highlights (e.g. MNIST digits) on a busy background

How to use:

1. Construct a Flow object with the input as source:

Flow(numpy_array_2d) or Flow("path/to/image.png")

2. Call consecutive image operations on the returned product. Each image operation returns the modified flow object, designed to make chaining function calls easier

Ex: to find the binary threshold of an image, dilate the image twice, and flip along the x axis: Flow(np_array).thresh(150).dilate(2).flip(axis=0)

3. Return the image as a PIL gresycale image:

Flow(np_arrray).im()

Functions supported:

• flip(axis:int): flip image along axis.
• thresh(thresh_value): threshold image at a certain value. I.e. all entries > value return 1, all entries < value return 0.
• thresh_scale(thres_value): all pixels below the value become 0, all pixels above the value are scaled from 0 to original value.
• std(mean, std_deviation): standarize image with mean and std_deviation: (pixel[i] - mean) / std_deviation
• pos(): convert every entry to positive (>= 0)
• erosion, dilation, closing, opening: perform the binary morphological operation. Optional arg itters determines how many times to repeat operation. Optional arg selem specifies a binary nxn np.ndarray which specifies neighbors considered in funciton.

Returning functions:

These functions end the chain of methdods and return a value

• im(): returns a PIL image (data must range from 0->1)
• imn(): returns a normalized PIL image
• imz(): returns a zoomed (256x256 pixesl by defualt) and unnormalized PIL image
• imnz(): returns a zoomed and normalized PIL image
• array(): returns the raw data in the array

Todo list:

• Implement parallelization using joblib.
• Display multiple images on the same line using matplotlib.pyplot to draw images
• Implement additional greyscale operations
• Improve support for multichannel operations