junliangliu / Captcha
基于CNN的验证码整体识别
Stars: ✭ 125
Programming Languages
python
139335 projects - #7 most used programming language
Projects that are alternatives of or similar to Captcha
Keras transfer cifar10
Object classification with CIFAR-10 using transfer learning
Stars: ✭ 120 (-4%)
Mutual labels: cnn, cnn-keras
Cnn keras
CNN | Keras/PyTorch | CAPTCHA recognition(卷积神经网络、Keras/PyTorch框架、验证码识别)
Stars: ✭ 256 (+104.8%)
Mutual labels: cnn-keras, captcha
Vincent Ai Artist
Style transfer using deep convolutional neural nets
Stars: ✭ 176 (+40.8%)
Mutual labels: cnn, cnn-keras
Object Localization
Object localization in images using simple CNNs and Keras
Stars: ✭ 130 (+4%)
Mutual labels: cnn, cnn-keras
Kaggle Web Traffic Time Series Forecasting
Solution to Kaggle - Web Traffic Time Series Forecasting
Stars: ✭ 29 (-76.8%)
Mutual labels: cnn, cnn-keras
Brdnet
Image denoising using deep CNN with batch renormalization(Neural Networks,2020)
Stars: ✭ 141 (+12.8%)
Mutual labels: cnn, cnn-keras
DLSS
Deep Learning Super Sampling with Deep Convolutional Generative Adversarial Networks.
Stars: ✭ 88 (-29.6%)
Mutual labels: cnn, cnn-keras
Image classifier
CNN image classifier implemented in Keras Notebook 🖼️.
Stars: ✭ 139 (+11.2%)
Mutual labels: cnn, cnn-keras
Benchmarking Keras Pytorch
🔥 Reproducibly benchmarking Keras and PyTorch models
Stars: ✭ 346 (+176.8%)
Mutual labels: cnn, cnn-keras
Personality Detection
Implementation of a hierarchical CNN based model to detect Big Five personality traits
Stars: ✭ 338 (+170.4%)
Mutual labels: cnn, cnn-keras
Screenshot To Code
A neural network that transforms a design mock-up into a static website.
Stars: ✭ 13,561 (+10748.8%)
Mutual labels: cnn, cnn-keras
Resnetcam Keras
Keras implementation of a ResNet-CAM model
Stars: ✭ 269 (+115.2%)
Mutual labels: cnn, cnn-keras
Hyperopt Keras Cnn Cifar 100
Auto-optimizing a neural net (and its architecture) on the CIFAR-100 dataset. Could be easily transferred to another dataset or another classification task.
Stars: ✭ 95 (-24%)
Mutual labels: cnn, cnn-keras
Aspect Extraction
Aspect extraction from product reviews - window-CNN+maxpool+CRF, BiLSTM+CRF, MLP+CRF
Stars: ✭ 117 (-6.4%)
Mutual labels: cnn
Unified Gesture And Fingertip Detection
A Unified Convolutional Neural Network Approach of Gesture Recognition and Fingertip Detection.
Stars: ✭ 121 (-3.2%)
Mutual labels: cnn
Yolo V3 Iou
YOLO3 动漫人脸检测 (Based on keras and tensorflow) 2019-1-19
Stars: ✭ 116 (-7.2%)
Mutual labels: cnn
DeepLab
机器学习、深度学习、量化投资,好用的工具、实用的工具,尽在DeepLab !
验证码识别
问题
通常验证码识别需要如下几个步骤:
(1)将整张图片切割为多个小图片,每个小图片包含一个字符
(2)训练模型,识别每个小图片中的字符
(3)将每个小图片的识别出的字符拼接为字符串作为整体识别结果
复杂一点的验证码往往各个字符不是均匀分布,难以准确切割。是否可以不切割图片直接识别其中的字符串?
思路
这其实是一个多标签分类问题,每个验证码图片有4个字符(标签),并且顺序固定;只要将卷积神经网络的最后一层softmax稍加修改就能实现多标签分类。
如下图所示,假设我们的验证码一共有4个字符,每个字符取26个大写字母中的一个;将卷积神经网络的输出层激活函数修改为sigmoid,输出层的[0-25]输出值对应第一个字符的onehot编码,[26-51]输出值对应第二个字符的onehot编码,[52-77]输出值对应第三个字符,[78-103]输出值对于第四个字符,并使用binary_crossentropy作为损失函数。
实验环境
keras theano cv2
训练集4000张图片,测试集2000张,每张图片大小20*80
网络结构
代码结构
generate.py 在GenPics目录下随机生成6000张验证码图片及其对应验证码字符串。
cnn_end2end_ocr.py 端到端验证码识别模型的训练、测试。
模型结果
Epoch = 60,Test Whole Accurate : 0.994
对于本文使用的较为简单的验证码,测试集整体正确率(4个字符都正确识别)99%以上。
参考:
Note that the project description data, including the texts, logos, images, and/or trademarks,
for each open source project belongs to its rightful owner.
If you wish to add or remove any projects, please contact us at [email protected].