Zhou Fugen

Edge Detection Based on Mathematical Morphology and Iterative Thresholding

Edge detection is a crucial and basic tool in image segmentation. The key of edge detection in gray image is to detect more edge details, reduce the noise impact to the largest degree, and threshold the edge image automatically. According to this, a novel edge detection method based on mathematic morphology and iterative thresholding is proposed in this paper. A modified morphological transform through regrouping the priorities of several morphological transforms based on contour structuring elements is realized first, and then an edge detector is defined by using the multi-scale operation of the modified morphological transform to detect the gray-scale edge map. Finally, a new iterative thresholding algorithm is applied to obtain the binary edge image. Comparative study with other morphological methods reveals its superiority over de-noising capacity, edge details protection and un-sensitivity to the shape of the structuring elements

Multi-modal image registration by mutual information based on optimal region selection

Mutual information (MI) has been commonly used in multi-modal image registration. In this paper, we presented an optimal region selection method for MI based multi-modal image registration. Firstly, image preprocessing and initial registration were applied on the two images. Then we applied two region selection procedures to improve the performance of MI-based registration. The first procedure is hierarchical region detection that selects regions with more structure information in each of the two images. In this procedure, the images were divided into sub-images progressively, and candidate image blocks with high usability were selected based on entropy. The second procedure is optimal matching region selection based on local registration using MI. In this procedure, the regions selected in the first procedure were registered with MI, then the results are used to choose optimal matching regions with high reliability. Finally, the optimal regions were used for the entire image registration by computing the MI of the regions. The experiment results showed that this method could improve the efficiency and accuracy for MI-based image registration.

A Novel Approach to Intra-operative Planning for Prostate Brachytherapy

This paper presents an efficient and stable algorithm to intra-operative planning for prostate brachytherapy. As a heuristic approach, the algorithm selects one seed at a step and iterates until predefined prostate volume is covered by prescribed dose. For each step, potential seeds are evaluated according to the ability to irradiate target while spare adjacent organs at risk. And its influence on dose homogeneity is also taken into account. Furthermore, a flexible mechanism is adopted to limit the number of used needles. The mechanism selects acceptable sub-optimal seed in existing needles instead of optimal seed that requires adding a new needle. The efficacy of the algorithm is evaluated on five clinical patient cases. Compared with state-of-the-art methods, this algorithm is efficient, stable and without the trial and error process of determining various parameters. These advantages make the algorithm suitable for intra-operative real-time treatment planning.

Mutual information based 3D registration of rat brain MRI time-series

A mutual information (MI) based registration method is introduced to align the rat brain tissues in MRI time-series, which is critical to the MRI tracer method for quantitative analysis of brain extracellular space (ECS). Specially, two works have been done to address the specific properties of contrast-enhanced MR time-series. First, the use of MI as similarity metric is validated by analyzing its robustness to local intensity variation caused by contrast enhancement. Second, an interactive segmentation method is incorporated into registration framework to extract the brain tissue in reference image. The resultant segmentation is used in MI computing to avoid the adverse effect of surrounding deformed tissue. Several experiments are conducted to evaluate the registration method qualitatively and quantitatively. The experimental results show that the proposed method has a high degree of accuracy and reliability, and is adequate to the task of 3D registration of rat brain MRI time-series.

Supervised image segmentation using learning and merging

The segmentation problem can be viewed as a learning and merging problem based on superpixels (image segments), which can incorporate a group of cues to guide the segmentation. So the proposed multi-label segmentation algorithm mainly consists of two stages: the learning stage and the merging stage. In the learning stage, Gaussian Mixture Models (GMMs) firstly learn color models for different components of objects. Based on the likelihood, we execute the alpha-expansion algorithm only once in order to alleviate the shrinking bias. The initial labels help determine whether a superpixel is too noisy, and the contour responses between superpixels can distinguish spurious boundaries. Those superpixels containing too much noisy pixels and spurious boundaries will be unlabeled. In the merging stage, unlabeled superpixels may have similar color information while differing in texture information. Therefore, they can be correctly classified by a novel region merging algorithm based on maximal similarity. In this way the advantages of features in different levels are enhanced by uniting them in different stages. Finally, the proposed method is evaluated on the Berkeley segmentation benchmark, the Graz benchmark and the Grabcut benchmark. Experimental results show that our method obtains the highest accuracy on the Graz benchmark, and the performance on other benchmarks can also be comparable or better than current leading algorithms.

A Modified FDK with Misaligned Parameters of Flat-panel Detector in Cone-Beam CT

FDK reconstruction algorithm is the most widely used for Cone Beam CT imaging system due to its simplicity and fast implementation. However, if misalignments exist in the detector, the original FDK reconstruction algorithm will result in deteriorated reconstruction quality because of the wrong correspondence of the voxels from the object and the pixels from the projection. One way is to apply rotation-based interpolation for each projection, which not only will cause blurring of the details due to the effect of interpolation but also is very time-consuming. In back-projection based reconstruction, the key factor is to correctly relate the voxels from the object and the pixels from the projections. Based on this point of view, we propose a method that will correctly find the correspondence of the voxels and pixels. With this method, reconstruction result can be obtained with good accuracy if all the misaligned parameters of the detector are known. And also, detail information can be better preserved compared with the rotation-based interpolation method. In the validation period, a simulation program of Cone-Beam CT was developed. In this program, all misaligned parameters of detector are able to be set. And the simulated result proves that our method is efficient and feasible although it suffers a little deterioration, and can further be used in online calibration of CBCT.

X-ray angiogram image enhancement in DSA with generalized fuzzy operator

It is believed that fuzzy set theory is a useful tool for handling the uncertainty associated with vagueness and/or imprecision. In this paper, by using a fuzzy entropy approach, a novel adaptive image fuzzy enhancement algorithm is presented. After a general discussion on fuzzy entropy, the concept of elementary entropy function of a fuzzy set is introduced. Using this mapping, the selection of the threshold value in image enhancement is beneficial to a certain extent. The second part of the paper investigates the applicability of the generalized fuzzy operator (GFO), which not only has a closing character and an automatic-adjusting character, but also has a transplant character to other enhancement arithmetic. One typical example is used for evaluation this algorithm at last, which is the X-ray angiogram image come from the Digital Subtraction Angiography (DSA).

GPU based acceleration architecture for image enhancement in spatial domain

In order to reduce the processing time of image enhancement in spatial domain, a GPU (Graphic Processing Unit) based acceleration architecture is proposed and implemented. With structured design method, computing model, data and algorithm resource which are indispensability in GPU computation are encapsulated, and computed directly in high performance with CUDA (Compute Unified Device Architecture). This architecture shields the configuration details of GPU computation and reduces repetitive work. In addition, new algorithms of enhancement in spatial domain could be added conveniently in the architecture. More importantly, the executing time of algorithms could be reduced 12–38 times than before in CPU, which is useful for the applications in real time system. For the neighborhood algorithms of image enhancement, a better solution of texture memory is used. Though this way, the time of executing algorithms could be reduced 36–135 times.