Quan Pang

Overlapped Cell Image Segmentation Based on Distance Transform

In micro cell images, several cells overlapped with each other often appears which causes difficulty to automatic cell estimation. How to separate overlapped cells to individual cells becomes the prerequisite of analyzing the micro cell images. In this paper, by introducing the theory of distance transform combining with watershed algorithm, a new algorithm of overlapped cell segmentation is proposed. This algorithm adopts distance transform to turn the position information of pixels to grey level information, and then adopts the watershed algorithm to realize the search and segmentation of the adhesion border of cell. The main characteristic of this method is that it can reduce the requisition of the quality of image, such as grey difference or noise level and so on. At the same time, it can calculate fast with high accuracy and steady result. So it is suitable for most overlapped cell images segmentation

A new unit layer linear rate control algorithm for H.264 based on PID controller

Abstract This paper presents a new unit layer rate control algorithm for H.264 combining PID controller with linear rate model. In various video coding standards, like MPEG-2, H.263, MPEG-4 and H. 264, it has been reported that the linear rate control algorithm can achieve more accurate and robust rate control. In fact, human is not only sensitive to spatial quality, but also to temporal quality. In order to have better tradeoff between spatial and temporal quality and obtain more consistent quality, the PID controller is inducted into unit layer linear rate control algorithm for H. 264. The scene changes are also effectively dealt with. Experimental results show that the proposed rate control algorithm can not only track target bit rate more accurately and achieve significantly smaller bit rate estimation error, like that of linear rate control algorithm, but also improve the temporal quality while keeping high spatial quality, and the quality of frames with scene changes can obviously be improved.

Fibre-optic Chemical Sensor Based on Characteristic Spectrum Recognition and its Application

In this paper, a characteristic spectrum recognition method based on artificial neural network is proposed to improve the sensitivity and selectivity of the chemical sensor. The requirements on the selectivity and the sensitivity of the chemical sensing materials can be lowered while the sensing function and sensing range raised effectively, which will be benefit to the practical realization and application of the chemical sensing system.

Texture Image Segmentation Based on Description Complexity

From fractal simulation theory, texture images can be reproduced by some texture sets through a nonlinear plural dynamical system. This paper generalizes the KC complexity measure, which is often used in analyzing the complexity of one-dimension time sequence into two-dimension image. The tests prove that the complexity description based on the KC complexity measure is effective. An improved measure method based on the spatial redundancy, is proposed to reduce the sensitivity to noises and to improve the robustness. Comparing with other usual algorithms of texture segmentation, the proposed algorithm has the advantages of less computation and better segmentation performance.

Image segmentation based on the local minium cross-entropy and quad-tree

With maximum entropy principle, satisfactory segmentation can be attained in dealing with the various sizes of objects. However, for some inhomogeneous images, due to the factors of inhomogeneous illumination, the global threshold cannot be used to segment all objects. On the basis of the current threshold algorithms and with the deduction of the relationships between entropy of the original set and ones of subsets, this article develops an image segmentation method based on local minimum cross-entropy, so to meet the requirements of inhomogeneous cell images. Moreover, the article presents a realization process of the algorithm that is combined with the quad-tree model, which has the advantageous of less computation and better segmentation effect, in comparison with other algorithms of adaptive threshold method.

A New Type of Photochemistry Instrumentation Based on Multi-sensor and Multi-spectrum

Up to now, the main obstacle in developing photochemistry instrumentation is the lack of satisfactory sensitivity and selectivity of chemical sensor. Many efforts have been made to solve this problem, for instance, by composing sensing array with different materials and data treatment algorithm based on artificial neural networks. These efforts, however, are quite limited due to the fact that the excitation signal on single or fixed frequency is not suitable for the sensors with different optical characteristics. In this paper, a new type of photochemistry instrumentation based on multi-sensor and multi-spectrum is proposed. The multi-sensor is composed of sensor array made by non-selective sensing materials and excited by a continuous spectrum. As the continuous spectrum covers a relatively wide frequency scope, the sensor array can describe different optical characteristics and create abundant sensing information. The adaptive sampling procedure of the spectrum and data processing algorithms is also given for the universal pattern recognition. Experimental results show that the proposed system not only works well on-line but also reduces the requirement of sensitivity and selectivity of the sensor. Therefore, the proposed system is helpful for the practical realization of photochemistry instrumentation.