图像分割是图像处理中的重要一步,然而图像容易受到空间光照等因素的影响使得图像灰度不均匀,而传统的基于区域的水平集方法在灰度不均匀时分割效果较差.为了提高图像分割效果,将空间聚类信息引入到传统的水平集方法中可以取得较好的分割效果.但是水平集方法计算量大、实时效果差,限制了其应用范围.利用NVIDIA的CUDA平台将算法并行实现,有效地提高了算法的运行效率.文中利用CUDA的线程模型、共享内存等存储器将聚类中心计算、曲线演化计算和偏置域计算并行实现,并在灰度不均匀的CT图像和植物叶片图像上进行分割实验,相比于CPU上的串行实现,并行后的分割速度得到明显增加.%Image segmentation is an important step in image processing.but,since spatial variations in illumination and imperfections of imaging devices and other reasons often leads to image intensity inhomogeneity,which make the result of region-based level set model for image segmentation is not ideal.A level set method combination with spatial clustering information can segment the image which has intensity inhomogeneity very well.Due to its higher computing complexity,Algorithm can not achieve real-time requirements,It limits the application in practice.An efficient parallel implementation of intensity inhomogeneity image segmention with CUDA make the speed of segmention faster.In the experiment,the estimation of the bias field,compute of the cluster center,compute of the evolving curve implicitly represented by the zero level set are implemented by CUDA use the thread hierarchy and the shared memory,the result show this method has faster segmentation speed compared with running on a CPU.Real-time has greatly improved.
展开▼