Research And Development On Microscope Image Mosaics System Of Cervical Cytology

Cervical cancer ranks second only to breast cancer as the most common malignant tumor among the woman all over the world, and its mortality once occupied the first place in the woman malignant tumor. For decades, the improvement of early cervical screening technique reduces incidence of mortality from cervical cancer obviously. However, multiple relevant data prove that incidence of morbidity from cervical cancer shows an up and younger trend. So the research work on diagnoses and treatment aimed at the early cervical cancer must be carried on in the future. Computer-assisted Cytology Test (CCT) system just advanced this transformation of the early cervical cancer screening technique from traditional manual screening to computer intellectualization. Meanwhile, it makes a great contribution to the increased sensitivity and the decreased false-negative rate.CCT detects early symptom through analyzing and recognising an array of microscope images of cervical cells, which are acquired by electron microscope scan. The limited visual field caused by high enlargement factor of electron microscope results in it can only collect part-image from entire smear each time. And the image analysis and processing system of current CCT has to include image mosaics function. It is used to combine several ordered and mutually overlapped local microscope images of cells to one piece of image, which has a larger visual field and higher resolution so that an entire cervical cell smear can reappear. By now, there have been many kinds of image mosaicing algorithms. And their common goal is to improve the mosaicing precision and shorten the mosaicing time. But no algorithm can be called the optimal solution.A microscope image mosaics system designed for cervical cytology is presented in this paper, and developed on C++ Builder 6.0 under Windows XP. This software contains several function modules, such as image distribution and adjustment, image display, image mosaics, result saving , etc.. The image mosaics is introduced according to the following steps: image pre-processing, image registration and image composition. Of above, the first step includes image gray level transform, image geometric correction, image smooth denoising and so on. In image registration, a improved feature-based image registration algorithm is given for mosaicing efficiency increasement. And then, put forward a proposal that applying the weighted smooth method based on gray value to color image to finish the seamless image mosaics in image composition. All in all, after studying and comparing many matured image mosaicing algorithms, domestic and foreign, we make an effort to find a balance between time and accuracy of mosaicing. Finally, better algorithms are described, which can combine several local image to one well. Not only can it maintain a certain mosaicing speed of running, but also reduce the rate of error- matching in some degree.The mosaicing algorithm advanced in this thesis is proved that it can complete a high quality and rapid image mosaics. And it can help the doctors observe the configuration of cells more directly and easily to obtain a right examination result, lowering the error rate. This algorithm also can apply to other 2D image mosaics so long as it is improved further. However, with the number of images increasing, we hope to weaken the impact from error cumulation, and optimize the mosaicing algorithm of information-lacking area.