This paper is an extended version of the paper [11]. We introduce a method for the automatic detection of tree rings and the subsequent measurement of annual ring widths in untreated cross-section images. The goal of this approach is to provide additional insights into wood quality. The method consists of two main parts. The first one involves detecting tree rings using directional filters combined with an adaptive refining process, which extracts the rings from radial information at various angles around the tree pith. The second part involves creating a confidence map using polar quad-tree decomposition, enabling the identification of relevant image areas for measuring tree ring widths. The method is evaluated on two public datasets and compared with three recent neural networks. More precisely, one dataset is composed of raw wood cross-section images, while the other contains treated wood cross-section images. Two neural networks, DeepLabV3+ [15] and SegFormer [32], have been fine-tuned, whereas Segment Anything Model [17] have been applied in a zero-shot way, meaning without fine-tuning. We evaluated our method with and without the confidence map algorithm. The obtained results demonstrate a good performance in both detection and measurement with precision comparable to state-of-the-art neural networks used for segmentation.