Abstract: Photic extremum lines (PELs) are view-dependent, object-space feature lines that characterize the significant changes of surface illumination. Though very effective for conveying 3D shapes, PELs are computationally expensive due to the heavy involvement of the third- and fourth-order derivatives. Also, they require the user to manually place a few auxiliary lights to depict the model details, which is usually tedious work. To overcome these challenges, we present a novel computational framework that improves both the speed and quality of PELs. First, we derive a simple, closed-form formula of gradient operator such that various orders of derivatives can be computed efficiently and in parallel using graphics processing units (GPUs). The GPU-based PEL extraction algorithm is one order of magnitude faster than the original one. Second, we propose to extract PELs from various non-photorealistic shadings that not only depict the overall shape but also bring out the details at different frequencies simultaneously. As a result, the user can easily control the relative emphasis to different scales and obtain the desired line drawing results. We demonstrate the improved PELs on a wide range of real-world objects.