Elsevier

Pattern Recognition

Volume 43, Issue 4, April 2010, Pages 1199-1206
Pattern Recognition

Active contours driven by local image fitting energy

https://doi.org/10.1016/j.patcog.2009.10.010Get rights and content

Abstract

A new region-based active contour model that embeds the image local information is proposed in this paper. By introducing the local image fitting (LIF) energy to extract the local image information, our model is able to segment images with intensity inhomogeneities. Moreover, a novel method based on Gaussian filtering for variational level set is proposed to regularize the level set function. It can not only ensure the smoothness of the level set function, but also eliminate the requirement of re-initialization, which is very computationally expensive. Experiments show that the proposed method achieves similar results to the LBF (local binary fitting) energy model but it is much more computationally efficient. In addition, our approach maintains the sub-pixel accuracy and boundary regularization properties.

Introduction

The active contour models (ACM) [1], which are based on the theory of surface evolution and geometric flows, have been extensively studied and successfully used in image processing. The level set method proposed by Osher and Sethian [2] is widely used in solving the problems of surface evolution. Later, geometric flows were unified into the classic energy minimization formulations for image segmentation [3], [4], [5], [9], [10], [11], [16], [17], [23]. Generally speaking, the existing ACM methods can be classified into two types: edge-based models [1], [2], [3], [4], [6], [9], [14], [19], [20], [23] and region-based models [5], [7], [8], [10], [11], [16], [17], [18], [21]. Each of them has its own pros and cons.

The edge-based models utilize image gradient as an additional constraint to stop the contours on the boundaries of desired objects. Usually, a stopping function is used to attract the contours to the desired boundaries. In order to enlarge the capture range of the force, a balloon force term is often incorporated into the evolution function, which controls the contour to shrink or expand. However, it is difficult to choose a proper balloon force. Either a too large or too small balloon force will result in undesirable effects [10].

Region-based models utilize the image statistical information to construct constraints, and have more advantages over edge-based models. First, they do not use the image gradient, and can successfully segment objects with weak boundaries or even without boundaries. Second, the initial contour can start anywhere in the image, and the interior contours can be automatically detected. One of the most popular region-based models is the C–V model [5], which has been successfully used in binary phase segmentation with the assumption that each image region is statistically homogeneous. However, the C–V model does not work well for the images with intensity inhomogeneity. Vese and Chan extended their work in [17] to utilize multiphase level set functions to represent multiple regions. These models are called the piecewise constant (PC) models. Nonetheless, both the C–V and the PC models have the drawback described above.

In order to segment images with intensity inhomogeneities, Vese and Chan [17] and Tsai et al. [16] proposed two similar models, which are called piecewise smooth (PS) model. However, these methods are computationally inefficient. More discussions above the properties of PC model and PS model can be found in [10], [11]. Li et al. [10], [11] proposed the LBF (local binary fitting) model, which utilizes the local image information as constraints, can well segment objects with intensity inhomogeneities. Furthermore, LBF model has better performance than PC and PS models in segmentation accuracy and computational efficiency.

In this paper, we propose a novel ACM model that can be used to segment images with intensity inhomogeneities. We utilize the local image information to construct a local image fitting (LIF) energy functional, which can be viewed as a constraint of the differences between the fitting image [10], [11] and the original image. Furthermore, a novel method is used to regularize the level set function by using Gaussian kernel filtering after each iteration. In addition, re-initialization is not needed in the proposed method. The complexity analysis and experimental results show that the proposed method is more efficient than the LBF model, while yielding similar results.

The rest of the paper is organized as follows. In Section 2, we review some classic models and indicate their limitations. Section 3 describes our model and its variational formulation. In Section 4, we validate our method by various experimentations on synthetic and real images. Conclusion is made in Section 5.

Section snippets

The Mumford and Shah (MS) model

In [8], Mumford and Shah formulated the image segmentation problem as follows: find an optimal piecewise smooth approximation function u of image I, which varies smoothly within each sub-region Ωi of image domain ΩR2, and rapidly or discontinuously goes across the boundaries of Ωi. They proposed the following energy functional:EMS(u,C)=Ω(u-I)2dx+μΩ/C|u|2dx+v|C|,xΩwhere |C| is the length of the contour C, μ,v0 are fixed parameters.

The unknown set C and the non-convexity of the above energy

LIF model and its variational level set formulation

A local fitted image (LFI) formulation is defined as follows:ILFI=m1Hε(φ)+m2(1-Hε(φ))where m1 and m2 are defined as follows:{m1=mean(I({xΩ|φ(x)<0}Wk(x)))m2=mean(I({xΩ|φ(x)>0}Wk(x)))where Wk(x) is a rectangular window function, e.g. a truncated Gaussian window or a constant window. In our experiment, we choose a truncated Gaussian window Kσ(x) with standard deviation σ and of size 4k+1 by 4k+1, where k is the greatest integer smaller than σ. Similar segmentation results can be achieved if

Experimental results

Our algorithm is implemented in Matlab 7.0 on a 2.8-GHz Intel Pentium IV personal computer. In this section, we apply our method to synthetic images and real images of different modalities, and use the same parameters ρ=1, ε=1, ς=0.45, n=3 and time-step Δt=0.025. Parameter σ is chosen by experience according to the images The Matlab source code of the proposed algorithm can be downloaded at http://www.comp.polyu.edu.hk/∼cslzhang/code/LIF.zip.

Conclusions

In this paper, we proposed a novel active contour model driven by local image fitting (LIF) energy. The proposed LIF energy functional has less computational complexity than the local binary fitting (LBF) energy functional. A novel level set method based on Gaussian filtering was used to implement our variational formulation, and the experimental results revealed that it is not only robust to prevent the energy functional from being trapped into local minimum, but also capable of keeping the

About the Author—KAIHUA ZHANG was born in Rizhao, China, in 1983. He received his B.S. degree in science and technology of electronic information from Ocean University of China (OUC) in 2006 and master degree in signal and information processing from the University of Science and Technology of China (USTC) in 2009. Currently he is a research assistant in the department of computing, The Hong Kong Polytechnic University. His research interests include pattern recognition and image processing.

References (24)

  • S. Osher et al.

    Fronts propagating with curvature dependent speed: algorithms based on Hamilton–Jacobi formulations

    Journal of Computational Physics

    (1988)
  • H.-K. Zhao et al.

    A variational level set approach to multiphase motion

    Journal of Computational Physics

    (1996)
  • M. Kass et al.

    Snakes: active contour models

    International Journal of Computer Vision

    (1988)
  • V. Caselles, R. Kimmel, G. Sapiro, Geodesic active contours, in: Processing of IEEE International Conference on...
  • V. Caselles et al.

    Geodesic active contours

    International Journal of Computer Vision

    (1997)
  • T. Chan et al.

    Active contour without edges

    IEEE Transaction on Image Processing

    (2001)
  • G.P. Zhu et al.

    Boundary-based image segmentation using binary level set method

    SPIE OE Letters

    (2007)
  • J. Lie et al.

    A binary level set model and some application to Mumford–Shah image segmentation

    IEEE Transaction on Image Processing

    (2006)
  • D. Mumford et al.

    Optimal approximation by piecewise smooth function and associated variational problems

    Communication on Pure and Applied Mathematics

    (1989)
  • C.M. Li, C.Y. Xu, C.F. Gui, M.D. Fox, Level set evolution without re-initialization: a new variational formulation, in:...
  • C.M. Li, C. Kao, J. Gore, Z. Ding, Implicit active contours driven by local binary fitting energy, in: IEEE Conference...
  • C. Li et al.

    Minimization of region-scalable fitting energy for image segmentation

    IEEE Transactions on Image Processing

    (2008)
  • Cited by (0)

    About the Author—KAIHUA ZHANG was born in Rizhao, China, in 1983. He received his B.S. degree in science and technology of electronic information from Ocean University of China (OUC) in 2006 and master degree in signal and information processing from the University of Science and Technology of China (USTC) in 2009. Currently he is a research assistant in the department of computing, The Hong Kong Polytechnic University. His research interests include pattern recognition and image processing.

    About the Author—HUIHUI SONG was born in Liao Cheng, China, in 1986. She received her B.S degree in science and technology of electronic information from Ocean University of China (OUC) in 2008. Currently she is a M.S. candidate of University of Science and Technology of China (USTC). Her research interests include pattern recognition and image processing.

    About the Author—LEI ZHANG received the B.S. degree in 1995 from Shenyang Institute of Aeronautical Engineering, Shenyang, P.R. China, the M.S. and Ph.D degrees in Electrical and Engineering from Northwestern Polytechnical University, Xi’an, P.R. China, respectively, in 1998 and 2001. From 2001 to 2002, he was a research associate in the Department of Computing, The Hong Kong Polytechnic University. From January 2003 to January 2006 he worked as a Postdoctoral Fellow in the Department of Electrical and Computer Engineering, McMaster University, Canada. Since January 2006, he has been an Assistant Professor in the Department of Computing, The Hong Kong Polytechnic University. His research interests include image and video processing, biometrics, pattern recognition, computer vision, multisensor data fusion and optimal estimation theory, etc. Dr. Zhang is an associate editor of IEEE Transactions on Systems, Man and Cybernetics, Part C.

    View full text