Optimized FPGA Implementation and Synthesis of Image Segmentation Techniques

Abstract

Image segmentation is the process of dividing the image into its constituents such as regions and objects. Edge detection algorithms detect the edge pixels, which are the edges of the image in which the pixels' intensity changes abruptly. The techniques namely Sobel, Prewitt, and Robert are the prominent gradient-based edge detection algorithms that are operated on the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$3\mathrm{x}3$</tex> and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$2\mathrm{x}2$</tex> kernels in the X and Y directions. This is performed in two different phases - python and Verilog(hardware descriptive language) for the synthesis on Basys 3 Artix 7 FPGA. In the first phase, the input image goes through the grayscale inversion for the gradient-based edge detection (Sobel, Prewitt, and Robert) operator. Later on, it goes through an edge detection algorithm that determines the edges of the entire image. The final image is then compared based on different input parameters like PSNR, MSE, and execution time required. In the second phase, the input grayscale image is passed through the line buffers to store the input pixels for further processing of the operators. The <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$3\mathrm{x}3$</tex> 3 and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$2\mathrm{x}2$</tex> kernels are pre-defined based on the operator for convolution and are then fed to the output buffer. The final images were then compared with the same input parameters. The proposed approach in the first phase was performed with no pre-built functions in python and the second phase was performed in Verilog. The proposed approach is expected to suggest a highperformance operator.