Deep Learning-based Brightness Adaptive Instance Segmentation Using CLAHE

This study proposes a brightness-adaptive instance segmentation algorithm utilizing CLAHE (Contrast Limited Adaptive Histogram Equalization) to address domain shift issues that degrade object segmentation performance in indoor environments. The proposed algorithm employs a BAE (Brightness Adaptive Equalizer) module based on CLAHE in the YUV color space to adjust contrast and enhance input data quality. The algorithm enhances recognition accuracy by integrating the YOLOv8 object recognition model with an exception-handling structure. Furthermore, the algorithm’s effectiveness is validated by comparing brightness distributions between the training and test datasets. The performance is quantitatively evaluated using metrics such as precision, recall, and mean average precision. Experimental results demonstrate that the proposed method mitigates performance degradation from domain shifts and enhances accuracy in different lighting conditions. This work enhances object recognition and segmentation in challenging lighting scenarios.