Study On Medical Multispectral Image Demosaicking And Segmentation

The clinicians usually measure the size of the wounds by using a millimeter scale tool and perform visual evaluation of the injury when perform wound assessment in traditional way,which would occupy excessive medical resources.Much more information of different rhagiocrine cells around the wounds can be obtained from multispectral images in more than three spectral dimension,and it can realize accurate descriptions of the wound.However,limited accuracy of multi-spectral image reconstruction and segmentation algorithms and the problem brought from processing of multi-spectral images contains noises restrict the applications of multi-spectral imaging technique in automatic wound assessment.To solve the problem that current image mosaic reconstruction algorithm cannot perform adaptive adjustments and process multispectral images contained noises,mosaic image reconstruction algorithm and adaptive distance switching filter based on noise reduction are proposed in this thesis;to solve the false estimation problem brought by the wound shape,photography condition and point of view,a semi-automated segmentation method based on graph theory is proposed;Based on these novel algorithms,an online diagnosis system with the function of multi spectral image acquisition and recognition is realized..The main contents of this paper are as follows:(1)Aiming at some demosaicing algorithms cannot make adaptive adjustment according to original images,a novel demosaicing algorithm based on multi-scale adaptive distance is proposed: firstly,a Wiener filtering model is used to convert the mosaic problem into the image restoration problem;then,correlation matrix,the known spectral information and unknown spectral information are estimated by using spatial distance,spectral distance the distance and gradient;the Wiener filter coefficient matrix is decomposed into a product of parameter matrix and correlation matrix,which changes the problem of filtering coefficients solving into system parameter solution;product filter coefficient matrix improved from correlation matrix of system parameters and the original image with known spectral information change adjustment,which realizes the adaptive multi spectral image demosaicing reconstruction.The experimental results show that,better results are achieved in multi spectral image data sets and wound public datasets,compared to mosaic algorithm and linear interpolation algorithm,three interpolation and two tree edge sensing algorithm based on multi-scale adaptive distance: zipper error substantially is reduced and clear imaging results in peak signal-to-noise ratio are obtained;signal-to-noise ratio were increased by 3.5 and 3.261 compared to the edge based sensing algorithm of the two fork tree,i.e.the unit pixel gray error is reduced by 33% and 31%;also,the proposed algorithm is stable and has similar performance in real datasets.(2)As the traditional demosaicing algorithm cannot process of multi spectral image with noises,two different switching median filtering techniques based on adaptive demosaicing algorithm with different distance weight scale are proposed to process the original image and to reduce noise of demosaicing results: first,neighborhood ballot is used to detect noise;mean and standard deviation of non-noise pixel in the neighborhood are used to define the threshold;all these are employed to carry out the second level detection of noise based on the first level detection;noise in the original image now can be processed successfully.second,a new type of demosaicing algorithm based on switching vector median filter is proposed to carry out post processing of the multispectral images;a proposed smooth area detection model is used to improve the similarity of vector median filter set pixels,which can solve the problem that the fine line texture of the image is treated as noise when processed with traditional algorithms.The experimental results show that the proposed two kinds of multi spectral image denoising algorithm in noisy environment outperform the original image containing noise reduction algorithm.(3)An improved Grab Cut chronic wound multi spectral image segmentation algorithm is proposed in this paper to solve the weak adaptive ability problem of automated segmentation method caused by the shape of the wound,the picture interference conditions,point of view and other factors The proposed method can get satisfactory segmentation results with less operations.The proposed algorithm can change the loss function by using foreground / background model edge information of the image and make it more sensitive to the edge;it also can improve the modeling accuracy by using spectral vector to measure the difference between pixels and transform the energy loss function.The experimental results show that the proposed algorithm needs less interactive operations and achieve better accuracy and recall rate,F-SCORE has achieved good results.Compared to the traditional Grabcut algorithm,the segmentation of complex shape of the wound of the proposed algorithm in this paper were increased by 7.8%,3.7% and 6.2%.Performance degradation of the proposed algorithm is weak,which indicate that the proposed algorithm has good adaptability.(4)Based on the above work,an online wound evaluation system with the function of mosaic filter array multi spectral imaging and multi spectral image segmentation is realized,which includes wound multi spectral imaging segmentation and multispectral image collection: according to the mosaic filter array multi spectral imaging,organization structure and hardware design the correlation between the performance and hardware through rate,quantum effects are described.The results show that the proposed system can obtain the effective results of the wound segmentation,and can obtain a more comprehensive description of the wound state compared with the color image evaluation technique.In summary,the online system with the functions of multi spectral image acquisition and recognition is realized based on a series improvements of multispectral image reconstruction and segmentation algorithm,which will generate important scientific and application meaning.