Face Classification between Dog(🐶) and Cat(😺) with accuracy of 97.02%

Import's required

from sklearn.svm import SVC
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split
from sklearn.pipeline import Pipeline
from sklearn.metrics import classification_report
from sklearn.metrics import confusion_matrix
import matplotlib.pyplot as plt
import seaborn as sn
import numpy as np
import pandas as pd
from joblib import dump
import cv2
import pywt
import os

Confusion Matrix(cat-0 & dog-1)

Method

• Firstly we need to detect the face of cat or dog on the given image and fetch it from the defined function by using CascadeClassifier of cat and dog

  cat_cascade = cv2.CascadeClassifier("./haarcascade_frontalcatface_extended.xml")
  dog_cascade = cv2.CascadeClassifier('./haarcascade_frontalface_alt2.xml')
  
  def detect_faces(img, cascade):
      gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
      faces = cascade.detectMultiScale(gray, scaleFactor=1.04, minNeighbors=4, minSize=(30, 30))
      return faces
  

• After fetching it we need to import the data sample into Jupyter notebook or any other IDE, then we need to do Wavelet Transformation of that image

  

• Then we need to Vstack or vertically stack the image for accurate measures

  X, y = [], []
  c_list = os.listdir('./cat_train')
  d_list = os.listdir('./dog_train')
  for i in c_list:
     img = cv2.imread(f'./cat_train/{i}')
     img_har = w2d(f'./cat_train/{i}','db1',5)
     combined_img = np.vstack((img.reshape(200*200*3,1),img_har.reshape(200*200,1)))
     X.append(combined_img)
     y.append(0)
     
  for i in d_list:
     img = cv2.imread(f'./dog_train/{i}')
     img_har = w2d(f'./dog_train/{i}','db1',5)
     combined_img = np.vstack((img.reshape(200*200*3,1),img_har.reshape(200*200,1)))
     X.append(combined_img)
     y.append(1)
  

• Then finally we get the features(x & y).

• AFter it just an Standard Scaling and Model building task remains.

• You can use diffrent models such as Ramdom Forest,KNN,Decision Tree etc... (I prefer SVM due to it's Kernal's)