Performance Improvement Of Viola-Jones Using Slicing Aided Hyper Inference (SAHI) For Multi-Face Detection

Performing multi-face detection using Viola- Jones is a challenge to provide more accurate and faster detection results. Viola-Jones is able to provide fairly good detection results as one of the most popular face detection algorithms although it provides slower face detection times when using high-resolution digital images. This study will discuss improving the performance of the Viola-Jones algorithm by adding the Slicing Aided Hyper Inference (SAHI) method at the pre-processing stage by detecting objects in certain slicing areas, accelerating and increasing accuracy for face detection. The dataset used is in the form of a WIDER digital image file. The dataset is tested using an application built with the ACCORD library by adding the proposed method. The test was carried out in stages to compare Viola-Jones with the proposed method of SAHI using several scan block-sizes in terms of accuracy and detection process time. The test results show an increase in the average value of accuracy in the proposed method with a scan block of 128 pixels from 81.97% to 86.65% and a decrease in the detection process time slightly faster with a difference of 320 milliseconds from the Viola-Jones method.