Search

Superpixel generation is to cluster the pixels with similar features and plays an important role for image segmentation. Conventional superpixel generation methods are more meaningful, however, the learning based method can generate the superpixels directly from the segments in the ground truth and achieve even better performance. In this work, an advanced superpixel generation algorithm that combines the advantages of conventional methods and modern neural network techniques is proposed. In addition to colors and locations, we find that the feature generated by neural networks also provide useful information for superpixel assignment. Simulations show that, with the proposed superpixels, a much more precise segmentation result can be achieved.

In wireless communication, it is inevitable that the signal is highly affected by the noise. For example, for the radar located in the sea shore, due to the effects of sea clutter and remote detection range, the signal to noise ratio (SNR) is only about 0~15 dB. In this manuscript, we develop an advanced noise determination and removal algorithm based on the deep learning method of the U-net. The U-net is a pixel-wise classification network and widely used in image segmentation. In this work, we find that it is also an effective way to determine whether a pixel in the time-frequency domain is the signal part or the noise part, even in the low SNR case. It is very helpful for reducing the noise effect and improving the accuracy of fundamental frequency analysis for radar signal processing.

Cell image segmentation is an important topic since it benefits for medical research and diagnosis. It is more challenging than other segmentation problems because cells have similar colors and their boundaries are not always obvious. In recent years, deep learning based methods, including the V-net, play an important role in image segmentation. In this paper, we proposed several techniques to further improve the performance of the V-net for cell segmentation. First, in addition to (i) cells and (ii) background, we add two labels: (iii) the edge between cells and background and (iv) the edges among cells. Since the properties of cell edges are quite different from those of cell bodies and background, adding these labels are helpful for improving the performance. Moreover, several morphology-based post-processing algorithms are applied for boundary refinement. With these techniques, the accuracies of cell segmentation can be much improved and the cells with poor contrast can still be well segmented.

Cell segmentation has attracted increasing interests since it is helpful for medical research and automatic diagnosis. In this paper, we proposed a method that can well segment cells by applying two stages of thresholds, a global one and an adaptive one, with morphology operations to overcome overlapping problems of cells.