This paper introduces a novel thinning algorithm, which thins the image by integrating parallel stages for selective pixel removal. Unlike traditional methods that apply fixed pixel removal rules to all pixels, our approach dynamically categorizes pixels into low-, medium-, and high-complexity regions based on local neighborhood features. The proposed method ensures that simple structures are efficiently thinned, while high-complexity regions (such as intersections and curves) are preserved, improving the quality of the resulting skeleton. To evaluate the proposed method, we compare our method with established techniques, including K3M and Zhang-Suen thinning, using benchmark datasets of binarized images. Results demonstrate that ParThin achieves superior connectivity preservation whilst maintaining efficient number of minutiae extracted from the Crossing Number algorithm. Thinning is a basic stage in image segmentation and hence feature extraction for object recognition. The implementation of our method will be made publicly available.