CNN/MNIST/keras/src/Feature Extraction.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Understand extraction of features using convolution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from keras.datasets import mnist\n",
    "\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Initialize Variables"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load pre-stored MNIST training and test data\n",
    "(x_train, y_train), (x_test, y_test) = mnist.load_data()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Initialize input and filter value\n",
    "# Play around with custom image and filter values to understand feature extraction better\n",
    "\n",
    "image = x_train[0]\n",
    "kernel = [[2,-1,-1],[-1,2,-1],[-1,-1,2]]\n",
    "\n",
    "# Custom image value\n",
    "#image = [[1,0,0,0,0],[1,1,0,0,0],[1,1,1,0,0],[1,1,1,1,0],[1,1,1,1,1]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Convert list (if true) to array\n",
    "image = np.asarray(image)\n",
    "kernel = np.asarray(kernel)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Initialzie array to hold convolution output\n",
    "# Array size depends on size of kernel and image to be convolved\n",
    "\n",
    "row_img, _ = np.shape(image)\n",
    "row_ker, _ = np.shape(kernel)\n",
    "\n",
    "# --------------------------------------\n",
    "\n",
    "row_conv = row_img - row_ker + 1\n",
    "conv_op = np.zeros(row_conv * row_conv)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Function - Convolution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "def convolution(conv_op, image, dim):\n",
    "    \n",
    "    count=0\n",
    "\n",
    "    for i in range(dim):\n",
    "\n",
    "        for j in range(dim):\n",
    "\n",
    "            conv_op[count] = (image[i][j]*kernel[0][0]) + (image[i][j+1]*kernel[0][1]) + (image[i][j+2]*kernel[0][2])+ \\\n",
    "                (image[i+1][j]*kernel[1][0]) +  (image[i+1][j+1]*kernel[1][1]) + (image[i+1][j+2]*kernel[1][2]) + \\\n",
    "                (image[i+2][j+1]*kernel[2][1]) + (image[i+2][j+1]*kernel[2][1]) + (image[i+2][j+2]*kernel[2][2])\n",
    "\n",
    "            if conv_op[count] >= 2:\n",
    "                conv_op[count] = 1\n",
    "            else:\n",
    "                conv_op[count] = 0\n",
    "\n",
    "            count += 1\n",
    "            \n",
    "    return conv_op"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Function - Image Visualization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def visualize(image, title, plot):\n",
    "    \n",
    "    dim = image.shape[0]\n",
    "    \n",
    "    plt.subplot(1, 3, plot)\n",
    "    plt.xticks([0,dim//2,dim-1])\n",
    "    plt.yticks([0,dim//2,dim-1])\n",
    "    plt.axis([0, dim-1, 0, dim-1])\n",
    "\n",
    "    plt.imshow(image, cmap='gray')\n",
    "    plt.title(title)\n",
    "    #plt.gca().invert_yaxis()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Perform Convolution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "conv_op = convolution(conv_op, image, row_conv)\n",
    "conv_op = np.reshape(conv_op,(row_conv,row_conv))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Visualize - Convolution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAb0AAACvCAYAAABpRwBgAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f61aa0bba20>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure(1, figsize=(5,5))\n",
    "fig.text(0.3, 0.85, \"Convolution - Feature Extraction\", size = \"xx-large\")\n",
    "fig.subplots_adjust(left=0.125, bottom=0.1, right=1.25, top=1.1,\\\n",
    "                    hspace = 2, wspace=1)\n",
    "\n",
    "# ---------------------------------------------------------------------\n",
    "\n",
    "visualize(image, \"Input\", 1)\n",
    "visualize(kernel, \"Filter\", 2)\n",
    "visualize(conv_op, \"Output\", 3)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}