Research On Feature-based Registration Of Two-dimensional Gel Electrophoresis Images

Proteomics is a very important part of the post-genome era of research. As the core ofthe proteomics technology, two-dimensional gel electrophoresis technology plays animportant role in the development of proteomics. Two-dimensional gel electrophoresisimages are obtained by separating the proteins as spots on a gel based on the isoelectric pointand molecular weight of the proteins. Because the large number of proteins on the gel image,computer-based analysis of gel images is of great significance for proteomic analysis.Computer-based analysis of gel images mainly includes image preprocessing, detection andquantitative of the protein, and gel matching, etc. And protein matching is an importantcontent of gel image analysis. After getting two dimensional gel electrophoresis images, firstcompare the protein structure and characteristics according to the protein points in response tothe area of distribution and find out the differences of protein. Then analyze these differencesin protein points and provide evidence for drug research disease diagnosis, and givecontamination analysis. Image registration technology allows different protein spots and canquickly and accurately be extracted to provide a guarantee for the analysis of protein. In thispaper, registration techniques are studied by taking protein spot of the gel electrophoresisimage as the research object. The main research work and achievements are as follows:(1) The basic theory about protein spots matching in gel images was systematicallystudied, and some basic knowledge of image matching were elaborated in detail, such as thetype of image transformation, the basic process of image registration and the classification ofimage registration and other related knowledge(2) The algorithm of image registration based on mutual information and imageregistration algorithm based on harris operator were presented. Through the comparativeanalysis, we found that image registration based on mutual information algorithm can realizethe automatic registration and its registration precision is high. However, this method need tocalculate the mutual information required to all the pixels participate in every optimizationsearch to compute a mutualinformation. So this method has a larger amount of calculation andregistration speed is slow. Although the image registration algorithm based on harris operatoris simple and its registration speed is fast, due to the angle at the time of detection threshold set, missing information or spurious feature points occurs easily.(3) Combined with the registration method based on mutual information and thecharacteristics of the image registration method based on harris operator, SURF feature-basedimage registration algorithm was presented. On the one hand, it reduces the computation andimproves the matching speed. On the other hand, there will be no false feature points whenfeature points are extracted. And it improves the efficiency of registration. However, theresponse obtained in the calculation of the hessian matrix of the feature point is small and theextracted feature point is less. So it makes lower efficiency of registration. In order to solvethis problem, in the detection of SURF feature points the improved SURF algorithm isweighted with approximate hessian matrix determinant. It increases matching points andreduces the false matching points. Image registration algorithm based on SURF feature in thefeature point matching use euclidean distance as a similarity measure. Its calculation iscomplex and the registration speed is slow. The improved SURF algorithm uses manhattandistance instead of euclidean distance as the similarity measure in the feature point matching.It reduces the complexity of registration and improves the speed of registration. Themanhattan distance is weighted to ensure the robustness of registration.