Research On Auto-Focusing Method For An Automatic Urinary Sediment Analyzer

Auto-focusing is a crucial technology broadly applied in digital imaging systems and various optical instruments and devices. For fully automated optical instruments such as automatic urinary sediment analyzer, auto-focusing is one of the necessary fundamental functions. The accuracy of auto-focusing directly influences the quality of the obtained images, and subsequently the hardship and results of the following image processing tasks. Researches on auto-focusing are thus of significant theoretical and practical importance.In the review of this thesis, the development history of the auto-focusing technology is briefly summarized, a concise introduction on common auto-focusing techniques is given, the advantages of auto-focusing are commented, and the application background and the importance of the proposed research are explained.Existing focus measures based on image processing techniques such as methods using frequency domain information, entropy-based methods and derivative-operator-based methods are then introduced in details. Advantages and disadvantages of the methods are analyzed.With the conclusions drawn from the analysis on the disadvantages of the existing focus measures, two new focus measures are proposed in Chapter 3 and 4, respectively. The method given in Chapter 3 uses edge detection to extract the rough contours of the imaged objects, and the definition of the image is judged by the image gradients along the contours. In this way, the image gradients within object or background regions, which contribute to the focus measure in the existing methods, are discarded, and the new focus measure is thus more compatible with the human perception of the image definition.The other method described in Chapter 4 follows a similar idea. However, instead of using the time-consuming edge detection, the proposed focus measure first partitions the image into square sub-image blocks, and the most significant image gradient(s) of each block are obtained as the representative of the image gradients in the block, and the sum of these representative image gradients gives the final focus measure. Impacts of image gradients in object and background regions on the focus measure is also filtered out in this approach. Both new focus measures are experimented on real world images in comparisons with several existing focus measures, and the results show that the proposed methods are advantageous over the existing ones.The search of the optimal image definition during a focusing process is inspected in Chapter 5. Existing approaches such as Fibonacci searching, function approximation and hill-climbing algorithm are introduced. Considering the real-time requirements on the auto-focusing apparatus, a hill-climbing algorithm with variable step length is employed. Since the focus measures proposed in Chapter 3 and 4 are computationally expensive and the real-time requirement on auto focusing is difficult to fulfill by directly applying the new focus measures, a coarse-to-fine focusing strategy based on the variable step length hill-climbing method is given. In this strategy, a less accurate but very fast coarse focus measure utilizing the Roberts operator is first applied to estimate the approximate focusing range where the optimal definition is located. The more time-consuming but more accurate new focus measure is then used to do a finer search in this small range to determine the best focusing, and thus to obtain the optimal speed-accuracy performance.