Research On Dynamic Rang Expanding For Microscope Images And Its Application Of Intelligent Tubercle Bacillus Recognition

Because of the strong infectivity of tuberculosis, the eleventh national five-year plan proposed the standardization, normalization and automation request to the bacteriology inspection method. However, there are also some difficulties lies in the tubercle bacillus micro image automatic diagnosis such as the complex background, the light intensity sensitivity of TB target and the difficulty to gain its characteristic information by single frame image. So, from the aspect of gaining the source images of tubercle bacillus, an information fusion method based on multi-frame of the identical field was proposed in this paper which will be used to improve the information of sputum smear microscopic image. An intelligent system for tubercle bacillus examination was also proposed here, by which we can automatically identify and count the tubercle bacillus from the microscopic image of the smear. The paper primary coverage includes:continuously and automatic gaining of micro images and expansion of its dynamic range; segmentation and features extraction for tubercle bacillus targets; automatic diagnosis and counting for tubercle bacillus. Emphatically, this paper researches the dynamic range expansion of the microscopic images and TB targets extraction. Firstly, a micro image formation platform was build which is mainly constitutes by CCD, microscope, stepper motor and its controller. Additionally, in order to automatic focus the tubercle bacillus micro image and gain a sequence of images with different exposure, control software is compiled. Then, by selecting the direction between the lens and the effective examination region of the smear, the way of progressive scanning can be confirmable. After obtained a list of microscopic images of tubercle bacillus with different exposure, considering that Gamma correction can enhance the color images'detail, a pyramid high-dynamic image fusion algorithm based on color image Gamma correction was proposed. In the aspect of target segmentation, this paper creatively propounded a method, which combines the exposure time of the microscopic images and three different target extraction methods. These methods are: target extraction method for color images based on single layer preceptron; target extraction based on BP neural network; target extraction method based on the Euclidean distance relative to the central kernel. After got the target of the images of different exposure, target-based image information fusion technology was used for more effective extraction. Finally, in the intelligent recognition aspect, this paper proposed a small object and pseudo object exclusion algorithm and a set of characters which will be next used for tubercle bacillus recognition. Simultaneously, a corresponding BP neural network was designed, with which, the effective recognition and counting tubercle bacillus could be achieve and the experiments result shows the effectiveness of the system.