Lung cancer has become one of the leading causes of death in the world. Clear evidence shows that early discovery, early diagnosis and early treatment of lung cancer can significantly increase the chance of survival for patients. Lung Computer-Aided Diagnosis (CAD) is a potential method to accomplish a range of quantitative tasks such as early cancer and disease detection, In computer-aided diagnosis of lung disease, accurate and fast pulmonary parenchyma segmentation is the core step. Watershed algorithm is used in this paper to segment and extract lung parenchyma. To reduce over-segmentation, an improved watershed segmentation method which marks foreground and background is proposed. This method is based on watershed algorithm using “Sobel” operator on edge detection and then using mathematical morphology opening and closing operations and morphological reconstruction to mark the foreground. Extracting the local maximum associated with object will constitute the binary marker image. By testing different types of images, it proposes that the algorithm in this paper can be consistent with human visual characteristics and get more accurate, continuous object boundary. Compared with other watershed improvement methods, this proposed one requires less computational complexity, more simple parameters, and can effectively reduce over-segmentation.