Abstract:

As an essential part of image composition, digital image segmentation is supported by most of recent commercial painting systems. Since it is not easy to fully automate image segmentation, there have been many practical tools adopting semi-automated segmentation techniques with some guidance from users. Intelligent scissors[1], also called live wire[2], are one of highly interactive tools based on graph search over an entire image. Another method called live lane[2] localizes the search domain to give an interactive feedback. Based on the live wire, the live lane trades off the repeatability of segmentation for its time efficiency. In this paper, we present a novel image segmentation tool called enhanced lane which ensures both efficiency and repeatability. That is, our enhanced lane can extract objects from an image interactively with its efficiency comparable to that of the live lane while also keeping its repeatability comparable to that of the live wire.

Key references:

[1] Eric N. Mortensen and William A. Barrett. Intelligent Scissors for Image Composition. In Computer Graphics (SIGGRAPH กฏ95 Proceedings), pages 191-198, 1995.

[2] Alexandre X. Falcao, Jayaram K. Udupa, Supun Samarasekera, Shoba Sharma, Bruce Elliot Hirsch, and Roberto de A. Lotufo. User-Steered Image Segmentation Paradigms: Live Wire and Live Lane. Graphical Models and Image Processing, No. 60, pp.233-260, 1998.

Video description:

1. Introduction (el_brief.avi)

In this video, we present a novel interactive image segmentation tool called Enhanced Lane. Starting from the initial seed point planted by the user, Enhanced Lane constructs the path map in the local window centered at the seed point. As the user interactively steers the cursor along the target boundary, Enhanced Lane extends and updates the path map in the sequence of local windows to display the minimum-cost path from the seed point to the current cursor location. A new seed point is planted where the mimimum-cost path digresses from the target boundary.

2. Incremental Path Map Construction (el_mag.avi)

This video clip shows the path map being incrementally updated and extended along with the movement of the local window. The left image contains the target object to be segmented out by the enhanced lane and the right image shows the magnifed view of the local window moving along the target boundary. The initial path map is constructed as a minimum-cost path tree in the first window centered at the seed point. As the local window moves along with the cursor, the path map is updated and extended to the next window. In the magnified view, the red box indicates the current window, and the blue box indicates the previous window. The yellow lines indicates the minimum-cost path up to the current cursor position. At any time instance, the path map is constructed and updated only in the current local window. The domain of the current path map construction includes only the vertices whose costs are higher than the minimum-cost vertex in the boundary of the previous window, which is indicated by the big cyan-colored dot in the magnified view. Since these vertices have possibilities to be updated to lower their costs, they should all be deleted from the minimum-cost path tree before the current path map update. That is why we see some backtracking in the magnified view.

3. Applications (el_app.avi)

With its high level of interactivity, high speed, and high repeatability, Enhanced Lane is capable of segmenting complex foreground objects from an arbitrary background of a noisy, low-contrasted image, regardless of its size. It is applicable to various fields including digital image composition, medical image analysis, key extraction, and so on. This video clip shows the use of Enhanced Lane for Tour Into the Picture (TIP) application. The foreground objects in the input image are interactively extracted with quick and easy motion. The extracted objects are placed in 3D space, to create walk-through images from interactively navigating into the scene.