Image Colorizer

Python program that converts a Black & White image to Colored. Uses various ML algorithms such as SVR to predict the color of each pixel.

Setup

1. Install Necessary Packages

   All the necessary packages can be installed by using pip:

   pip install -r requirements.txt

2. Download Dataset

   The dataset required to train the model can be created with the help of the web scraper included.

   python get-images.py

   The script is designed to go on an unlimited loop, terminate the script when necessary amount of images have been downloaded.

3. Train the Model

   The model can be easily trained by running the train-model script

   python train-model.py

4. Test the Model

   In order to test the model you need to pass a black and white as input:

   python test-model.py <image_name>

Process Explanation

Training The Model

1. Generating Sub squares -

   For a given image generate the following -

   • Array of subsquares

   • Array containing average U values for the sub squares

   • Array containing average V values for the sub squares

   1. Segmentation -

      Segmentation is done to change the representation of the image into something that is more meaningful and easier to analyze.

      Functions used -

      skimage.util.img_as_float()
      skimage.data.imread()
      skimage.segmentation.slic()

   2. Get YUV Values - From the RGB values that were found from the segmentation process calculate the YUV values using the conversion matrix.

      Formula Explanantion

      Use np.dot()to calculate dot product.

   3. Find n_segments -

      The N segment of the image will be the maximum value in the segments array plus 1.

   4. Computing Centroids -

      Create templates of U, V, centroids and point_count using np.zeros()

      Iterate through the segments array with the help of np.ndenumerate() and calculate each element in the arrays.

      Calculate Centroids, U and V with the help of point_count.

   5. Calculate the Sub Square -

      Create a template for the Sub Square.

      Calculate the sub square using fourier transform with the help of np.ftt.ftt2

2. Train Models -

   1. Training the base model -

      Using sklearn.svm.SVR() train the base models using SVM with constant values of C and Epsilon.

   2. Fitting the Model -

      Using the fit() function, fit the base model using X, U_L and V_L.

Checklist:

Modules Involved -

• get-images -- Design a web scraper to get images of landscapes that will end up becoming the data set.
• train-model -- Utilize Support Vector Regressions to segment and train on the dataset based on YUV colorspace.
• test-model -- Use multiple layers of SVR on test image then apply Markov Random Fields to find hidden chrominance values and generate an output.