AN EVALUATION OF CONNECTED COMPONENTS LABELLING USING GPGPU

Connected components labelling (CCL) is one of the basic steps in various image-processing applications such as image segmentation or recognition while performing surveillance or medical imaging. Due to increased demand for real-time processing; fast and efficient connected components labelling and analysis has become significant. It is a resource and time intensive process but if parallelized, it can be done efficiently with much higher performance. The implementation presented makes use of two-pass algorithm by exploiting parallelism provided by graphics card using Open Computing Language (OpenCL). Performance of GPU with varying CPU loads is examined. At the end, performance results are compared with different images and serial implementation on CPU due to its serial nature of execution.