Ying Gu

Efficient and robust image segmentation with a new piecewise-smooth decomposition model

Image segmentation is to separate the image domain into local regions. Implicit active contour models via the level-set method can provide smooth and closed contours. The major category is based on the Mumford-Shah image component decomposition model. Approximating the image domain by a set of homogeneous regions, segmentation based on the piecewise constant (PC) models fail in intensity inhomogeneous images. Requiring intensity smoothness in local regions, segmentation based on the piecewise smooth (PS) models can perform better for intensity inhomogeneity than those PC based. Existing PS-based techniques are inefficient and not robust to complicate intensity scenarios with noise. Here we introduce a new functional model by decomposing an image into three parts: PS components, PC components and noise components. We convexify the decomposition model by incorporating with relaxation techniques and optimize the PS components over the whole image domain. New numerical algorithms are also proposed to implement the above approaches efficiently. Numerical validation experiments show that the proposed approaches can achieve much faster, more robust and more accurate image segmentation than existing arts.

A new approach for multiphase piecewise smooth image segmentation

Image segmentation is to separate the image domain into sub-domains according to some optimization rule. Variational level-set image segmentation based on the Mumford-Shah model, together with level set formulation, leads to complicate implementation and expensive computational time. To overcome the difficulty in solving the complex coupled level set functions and handle any number of phases in the image, we improve the region-base active contour model in this paper. We formulate the image segmentation as a new multiphase minimization problem to handle images containing any integer phases of intensities. We also propose efficient numerical algorithms to solve such optimization problem without extra steps of convex relaxation. Numerical validation experiments on various images show better accuracy and robustness over existing methods.

A new Mumford-Shah type model involvinga smoothing operator for multiphase image segmentation

Variational image segmentation based on the Mumford-Shah model requires to solve the heat diffusion equations in evolving irregular subdomains. It brings about significant difficulties in efficient and accurate segmentation, especially, in multi-phase scenarios. In this paper, we propose a new Mumford-Shah type model involving a smoothing operator, acting a similar role as the diffusion process and avoiding complicated computation. We introduce two simplified numerical models based on the Gaussian operator and the bilateral operator respectively, and provide efficient numerical algorithms to solve both segmentation problems. Numerical validation experiments on various images show better accuracy and robustness over existing methods.

Robust laser stripe extraction using ridge segmentation and region ranking for 3D reconstruction of reflective and uneven surface.

Laser stripe extraction is essential to 3D reconstructions of objects by triangulation laser scanning. Industrial applications require robustness against the spurious points due to multiple reflections among reflective and uneven surfaces. Our approach is a three-stage algorithm, which first detects the potential laser stripe regions, then ranks the regions on their corrected scene irradiance densities, and finally conducts the column-wise peak detection using the region ranking. The proposed method is evaluated in terms of speed, accuracy and robustness on both simulations and a real application for steel groove inspection.

A Three-Color Coupled Level-Set Algorithm for Simultaneous Multiple Cell Segmentation and Tracking

High content computational analysis of time-lapse microscopic cell images requires accurate and efficient segmentation and tracking. In this work, we introduce “3LS”, an algorithm using only three level sets to segment and track arbitrary number of cells in time-lapse microscopic images. The cell number and positions are determined in the first frame by extracting concave points and fitting ellipses after initial segmentation. We construct a graph representing cells and the background with vertices and their adjacency relationships with edges. Each vertex of the graph is assigned with a color tag by applying a vertex coloring algorithm. In this way, the boundary of each cell can be embedded in one of three level set functions. The “3LS” algorithm is implemented in an existing coupled active contour framework (nLS) [1] to handle overlapped cells during segmentation. However, we improve nLS using a new volume conservation constraint (VCC) to prevent shrinkage or expansion on whole cell boundaries and produce more accurate segmentation and tracking of touching cells. When tested on four different time-lapse image sequences, the 3LS outperforms the original nLS and other relevant state-of-the-art counterparts in both segmentation and tracking however with a notable reduction in computational time.

Cascaded Shape Regression for Automatic Prostate Segmentation from Extracorporeal Ultrasound Images

Prostate segmentation from extracorporeal ultrasound (ECUS) images is considerably challenging due to the prevailing speckle noise, shadow artifacts, and low contrast intensities. In this paper, we proposed a cascaded shape regression (CSR) method for automatic detection and localization of the prostate. A sequence of random fern predictors are trained in a boosted regression manner. Shape-indexed features are used to achieve invariance against geometric scales, translation, and rotation of prostate shapes. The boundary detected by CSR is used as the initialization for accurate segmentation by using a dynamic directional gradient vector flow (DDGVF) snake model. DDGVF proves to be useful to distinguish desired edges from false edges in ECUS images. The proposed method is tested on both longitudinal- and axial- view ECUS images and achieves Root Mean Square Error (RMSE) under 1.98 mm (=4.95 pixels). It outperforms the active appearance model in terms of RMSE, failure rate, and area error metrics. The testing time of CSR+DDGVF is less than 1 second per image.

A Noise-Tolerant Algorithm for Robot-Sensor Calibration Using a Planar Disk of Arbitrary 3-D Orientation

In a 3-D scanning task, a robot-sensor system controls a robotic arm to move a laser sensor. In order to align the coordinate system of the robotic arm and laser sensor, prior calibration is required to derive the transformation between both coordinate systems. This paper proposes a new calibration method in three steps: manual data collection, sensing data calculation, and transformation solution. First, at least four data are required to be collected by the user. The sensing data are then calculated from the collected data and adopted to provide the desired transformation. The proposed algorithm has two features: arbitrary placement of planar disk and noise tolerant. Using a planar disk, the algorithm will automatically derive the angular relationship between the disk and the sensor plane, enabling arbitrary orientation placement. Noise tolerant is guaranteed by fitting ellipses during the sensing data calculation and using a single set of sensing data in transformation solution. Experiments and comparisons are given to demonstrate the efficiency of the proposed calibration algorithm.Note to Practitioners—This paper was motivated by the problem of calibrating a laser sensor and a positioning device (robot arm, CMM, etc.) in a robust and fast manner. Specifically, the calibration is to derive the transformation by aligning the sensors coordinate system to the positioning devices coordinate system. The proposed calibration procedure consists of two parts: manual data collection and automatic transformation calculation. During manual data collection, users only need to select four different data; whereby each data contains of two positions with the same orientation. Then, the desired transformation will be derived automatically. The calibration is designed in an efficient and robust way whereby: 1) data collection is done using a simple planar disk placed in arbitrary orientations; 2) minimum human interaction required; 3) tolerant to noise in the sensor data; and 4) easy implementation by following a proven and standard protocol.

Generalizing Mumford-Shah Model for Multiphase Piecewise Smooth Image Segmentation

This paper concerns multiphase piecewise smooth image segmentation with intensity inhomogeneities. Traditional methods based on the Mumford–Shah (MS) model require solving complicated diffusion equations evolving in irregular subdomains, leading to significant difficulties in efficient and accurate segmentation, especially in multiphase scenarios. In this paper, we propose a general framework to modify the MS model by using smoothing operators that can avoid the complicated implementation and inaccurate segmentation of traditional approaches. A detailed analysis connecting the smoothing operators and the diffusion equations is given to justify the modification. In addition, we present an efficient algorithm based on the direct augmented Lagrangian method, which requires fewer parameters than the commonly used augmented Lagrangian method. Typically, the smoothing operator in the general model is chosen to be Gaussian kernel, the bilateral kernel, and the directional diffusion kernel, respectively. Ample numerical results are provided to demonstrate the efficiency and accuracy of the modified model and the proposed minimization algorithm through various comparisons with existing approaches.This paper concerns multiphase piecewise smooth image segmentation with intensity inhomogeneities. Traditional methods based on the Mumford-Shah (MS) model require solving complicated diffusion equations evolving in irregular subdomains, leading to significant difficulties in efficient and accurate segmentation, especially in multiphase scenarios. In this paper, we propose a general framework to modify the MS model by using smoothing operators that can avoid the complicated implementation and inaccurate segmentation of traditional approaches. A detailed analysis connecting the smoothing operators and the diffusion equations is given to justify the modification. In addition, we present an efficient algorithm based on the direct augmented Lagrangian method, which requires fewer parameters than the commonly used augmented Lagrangian method. Typically, the smoothing operator in the general model is chosen to be Gaussian kernel, the bilateral kernel, and the directional diffusion kernel, respectively. Ample numerical results are provided to demonstrate the efficiency and accuracy of the modified model and the proposed minimization algorithm through various comparisons with existing approaches.

CHORD: Cascaded and a contrario method for hole crack detection

We propose a cascaded and a contrario hole crack detection (CHORD) method for defect inspection in digital images of turbine blade surfaces. This is the first time an automatic image processing-based method is proposed for such a task. It consists of two major steps: first, the appearance of holes is approximated by ellipses and cascaded pose regression is used to estimate the position and orientation of the holes; Second, we define hole cracks as geometrical structures and a contrario method is used to assess the meaningfulness of each crack. The model-based CHORD method fully considers the features of hole cracks including the characteristics of brightness, length, and orientation, and therefore can accurately detect cracks in the images. The threshold on the strength of cracks is determined automatically and the computational time is about five seconds for each image.

3D prostate segmentation from MRI images using modified rotational slice-based level set with non-uniform contour shrinking

Accurate prostate segmentation of 3D prostate images of different modalities plays a key role in image-guided biopsy and therapy of prostate cancer. Recently, an efficient rotational slice-based approach was proposed for transrectal ultrasound (TRUS) images, and accurate results were achieved. When we applied this method to MR images, we encountered several problems, including accumulation of error, and non-convergence due to absence of data. In this paper, we improved this work by proposing a modified rotational volume slicing method with a non-uniform contour shrinking mechanism, which gave better accuracy and robustness for 3D MRI prostate segmentation. The numerical experimental results demonstrated that the proposed method outperformed the original rotational slice-based approach.