Abstract : In this paper, we provide a layered non-photorealistic rendering (NPR) technique that automatically extracts the depth of field (DoF) shown in the picture and adjusts the degree of abstraction accordingly. We use an RGB channel to efficiently classify the DoF region anisotropically. Based on the DoF values, we abstract the color and adjust the thickness of the line. We use anisotropic DoF-based filtering to improve the abstraction quality by finding the blur region using cross-correlation filtering and anisotropically calculating the weight map. Our approach has greatly improved the quality of abstraction in terms of performance and design. The algorithm is also fast and simple to implement. Experimental results show well the characteristics and style of the DoF of the original photograph.

[paper]