Leaves image synthesis using generative adversarial networks with regularization improvement

Diversity of leaves form a special feature in a plant that can be done research such as image segmentation. However, the thing that is the main issue is the quantity of labeled data. Through image synthesis or image segmentation we are able to add leaf shape needed to use Generative Adversarial Networks (GAN). To train the GAN requires the choice of architecture, initialization parameters and more accurate selection as it often becomes a GAN challenge. Therefore appropriate regularization techniques are needed to address the problem. In the end we were able to segment the 3 leaf shape images using conventional GANs that we have modified using attention to the optimal regularizer parameters. Elastic Net or a combination of L1 and L2 regularizer that we tested on the second model gives error rate 0,105% for discriminator and 20,95% for generator.

[1]  Paulus Insap Santosa,et al.  Leaf Classification Using Shape, Color, and Texture Features , 2013, ArXiv.

[2]  Sebastian Nowozin,et al.  Stabilizing Training of Generative Adversarial Networks through Regularization , 2017, NIPS.

[3]  Hanno Scharr,et al.  ARIGAN: Synthetic Arabidopsis Plants using Generative Adversarial Network , 2017, bioRxiv.

[4]  Yoshua Bengio,et al.  Generative Adversarial Networks , 2014, ArXiv.

[5]  Alexei A. Efros,et al.  Generative Visual Manipulation on the Natural Image Manifold , 2016, ECCV.

[6]  Ole Winther,et al.  Autoencoding beyond pixels using a learned similarity metric , 2015, ICML.

[7]  Yoshua Bengio,et al.  Plug & Play Generative Networks: Conditional Iterative Generation of Images in Latent Space , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[8]  Yann LeCun,et al.  Deep multi-scale video prediction beyond mean square error , 2015, ICLR.

[9]  Claus Nebauer,et al.  Evaluation of convolutional neural networks for visual recognition , 1998, IEEE Trans. Neural Networks.

[10]  Luis Herranz,et al.  Heterogeneous Convolutional Neural Networks for Visual Recognition , 2016, PCM.

[11]  Yan Cui,et al.  Multi-phase Offline Signature Verification System Using Deep Convolutional Generative Adversarial Networks , 2016, 2016 9th International Symposium on Computational Intelligence and Design (ISCID).

[12]  Wojciech Zaremba,et al.  Improved Techniques for Training GANs , 2016, NIPS.

[13]  Yoshua Bengio,et al.  Mode Regularized Generative Adversarial Networks , 2016, ICLR.

[14]  Angel Domingo Sappa,et al.  Infrared Image Colorization Based on a Triplet DCGAN Architecture , 2017, 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW).

[15]  Jiajun Wu,et al.  Learning a Probabilistic Latent Space of Object Shapes via 3D Generative-Adversarial Modeling , 2016, NIPS.

[16]  Bernt Schiele,et al.  Generative Adversarial Text to Image Synthesis , 2016, ICML.

[17]  Pabitra Mitra,et al.  Generative Adversarial Learning for Reducing Manual Annotation in Semantic Segmentation on Large Scale Miscroscopy Images: Automated Vessel Segmentation in Retinal Fundus Image as Test Case , 2017, 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW).

[18]  Hanno Scharr,et al.  Leaf segmentation in plant phenotyping: a collation study , 2016, Machine Vision and Applications.

[19]  Soumith Chintala,et al.  Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks , 2015, ICLR.

[20]  Jie Yang,et al.  Brain MRI segmentation with patch-based CNN approach , 2016, 2016 35th Chinese Control Conference (CCC).

[21]  Hanno Scharr,et al.  Machine Learning for Plant Phenotyping Needs Image Processing. , 2016, Trends in plant science.

[22]  H. Zou,et al.  Addendum: Regularization and variable selection via the elastic net , 2005 .

[23]  Patrick O. Glauner.  Comparison of Training Methods for Deep Neural Networks , 2015, ArXiv.