Effect of Image Pre-Processing Techniques on Object Detection

Abstract: This work analyses how different image pre-processing techniques affect object detection results. Objects are detected after applying various processing methods, and the outputs are compared. Most existing studies focus on improving object detection models, but there is limited work analysing how individual pre-processing techniques influence detection results. YOLOv8 (You Only Look Once) is used in this study as it is a fast and reliable model pre- trained on the COCO (Common Objects in Context) dataset. The focus of this work is to observe how different techniques affect the number of detected objects and their corresponding confidence scores, which indicate how certain the model is about a detected object and its location. The effects of techniques such as greyscale conversion, histogram equalisation, contrast adjustment, blurring, and edge detection are analysed on real-world images and a subset of the COCO dataset. Approximately 20 images from the coco128 subset are used for controlled analysis. The evaluation is based on a relative comparison of detection outputs using detection count and confidence scores (above a threshold of 0.5). It is observed that different techniques perform differently depending on the image, and no single method consistently provides the best results. This study helps in understanding how pre-processing influences object detection behaviour and supports better selection of techniques based on the input image.

Keywords: Object Detection, YOLOv8, COCO Dataset, Image Pre-processing, Confidence Scores, Histogram Equalisation, Edge Detection