CAJ | 학술논문

针对传统的马尔科夫随机场算法中模型参数估计是全局的，及此算法描述非平稳SAR海冰图像是局限的，提出一种带有纹理保护的图像分割算法。该算法以区域为研究对象，首先利用分水岭分割算法对图像进行初始分割得到基本同质的区域，使该算法由像素水平提升到区域水平，这样能减少噪声对分割结果的影响。然后使用集成了纹理信息的空间语境模型和特征模型来描述对象函数，获得更稳定的模型参数估计，使得该算法具有描述局部行为的能力，改进了空间语境模型对图像非平稳性的适应性。通过对1幅合成图像和2幅真实合成孔径雷达海冰图像进行测试，将该算法与马尔科夫随机场算法和Gaussian混合模型算法比较，结果表明，该文算法优于上述2算法，在相同的场景内该文算法在产生平滑结果的同时也能保护纹理特征。
침대전통적마이과부수궤장산법중모형삼수고계시전국적，급차산법묘술비평은SAR해빙도상시국한적，제출일충대유문리보호적도상분할산법。해산법이구역위연구대상，수선이용분수령분할산법대도상진행초시분할득도기본동질적구역，사해산법유상소수평제승도구역수평，저양능감소조성대분할결과적영향。연후사용집성료문리신식적공간어경모형화특정모형래묘술대상함수，획득경은정적모형삼수고계，사득해산법구유묘술국부행위적능력，개진료공간어경모형대도상비평은성적괄응성。통과대1폭합성도상화2폭진실합성공경뢰체해빙도상진행측시，장해산법여마이과부수궤장산법화Gaussian혼합모형산법비교，결과표명，해문산법우우상술2산법，재상동적장경내해문산법재산생평활결과적동시야능보호문리특정。
This paper proposesd an image segmentation algorithm with texture preservation in view of the traditional Markov random field ( MRF ) image segmentation methods, the model of parameter estimation was global,and this algorithm was inadequate that described non-stationary SAR sea ice image was limited. Sea ice regions were researched as objects. The watershed algorithm was first used to generate primitive homogeneous regions. The impact of noise on the segmentation result could therefore be reduced in the space of regions instead of pixels. The proposed method incorporated texture information of feature model and spatial context model formulated the objective functions, which had some capability of describing local behaviors and could improve the spatial context model on its adaptivity to the non-stationarity of the image. In the traditional MRF approach, its models were stationary, with model parameters estimated globally. By testing on one synthetic image and two SAR sea-ice scenes, the algorithm of the paper is compared with Gaussian mixture model algorithms and MRF-based segmentation algorithms. The comparison indicated that the algorithm of this paper was more excellent than the above-mentioned two algorithms. The algorithm could simultaneously preserve texture feature and poduce smooth segmentation results in the same scene.