A Comparative Study on Input Format for Convolutional Neural Network based Head Detection

In this research, we study and evaluate the performance of CNN based head detection with various input image. We evaluate grayscale, CLAHE and saliency map format as inputs to our CNN model. We use INRIA dataset for training and testing data. For training data, we use image size of 30×20 pixels. The experimental result shows saliency map is a good input and adam is a good optimizer for our CNN model. The experiment is conducted using programming language python and openCV library.