Abstract

Image segmentation can extract valuable information from images and has very important practical significance. In this paper, the application of Convolutional Neural Network (CNN) in image processing is studied. Full Convolutional Network (FCN) is used to improve the accuracy of image feature extraction and Visual Geometry Group-16 (VGG-16) is improved. In order to further improve the accuracy of image local positioning, the FCN output and the Conditional Random Field (CRF) are combined to obtain the FCN-CRF segmentation model and the model is analyzed based on the Weizmann Horse data set as the experimental object. The result suggests that the FCN-CRF model in this paper can achieve accurate segmentation of images with an average precision of 86.48& and the average intersection-over-union of 72.67%, which is significantly higher than Support Vector Machine (SVM), K-means and FCN algorithms. Moreover, it only takes around 0.3 s to process each image. The algorithm in this paper is proved to be reliable by the result. This research provides theoretical support for the application of CNN in image segmentation processing, which is conducive to the further development of image segmentation technology.

中文翻译：

---

基于卷积神经网络的图像特征提取与分割处理

摘要

图像分割可以从图像中提取有价值的信息，具有非常重要的现实意义。本文研究了卷积神经网络（CNN）在图像处理中的应用。使用全卷积网络（FCN）来提高图像特征提取的准确性，并且改进了Visual Geometry Group-16（VGG-16）。为了进一步提高图像局部定位的准确性，将FCN输出和条件随机场（CRF）相结合以获得FCN-CRF分割模型，并以Weizmann Horse数据集为实验对象对该模型进行分析。结果表明，本文的FCN-CRF模型可以实现图像的精确分割，平均精度为86.48＆，平均相交重叠率为72.67％，明显高于支持向量机（SVM），K-means和FCN算法。此外，处理每个图像仅需约0.3 s。实验结果证明了该算法的可靠性。该研究为CNN在图像分割处理中的应用提供了理论支持，有利于图像分割技术的进一步发展。