Exploring the Impact of Image Filters in Facial Detection through Different Classification Algorithms
Raj Kishor Bisht, Daivik Mohan, Aadershi Mohan, Akshay Kumar Joshi, Yogesh Lohumi · 2024
In the present work, we have analyzed the effect of applying different filters in face detection through different classification methods. For the purpose, we have created different datasets: first dataset consists of 1600 different unfiltered images of eight students with 200 images of each student and second dataset consists of 8000 filtered images using a mix of filters with 1000 images of each student. Then we created five additional datasets, each dataset created by applying a single filter. Five filters namely blur, cut-out, mosaic, noise and shear are used for the present study. Six different algorithms CNN, Random Forest, SVM, Gaussian Naïve Bayes, KNN, and Logistic Regression were applied and results of unfiltered image dataset and filtered image dataset using a mix of filters have been compared. We found a little improvement in the accuracy in the case of KNN and CNN and a little decline in accuracy of other algorithms applied to filtered dataset created using a mix of filters. The significance of difference between the two results is checked through paired t-test. Based on paired t-test, we conclude that filters in images, in general, do not have much effect in face detection through different classification algorithms. The results of unfiltered dataset and individual filtered datasets show that there is a significant difference in results between the unfiltered dataset and blur filtered dataset, others being identical. This indicates that the filter ‘blur’ has some effect in face detection and other has no significant effect.