Machine learning/Computer vision/OCR

Hello, there, I am very happy to put my bid on your project. I am an expert of computer vision such as OCR and machine learning algorithms for 8+ years. If you see my profile, you can see my abilities and previous works. But I hope that we will discuss about all detail items of the project together asap, then I think we can get good result. I am looking forward your positive response. Regards. P.S. I don't check your sample invoices but I answer to your questions based on my previous experience. If invoice includes "total" or "purchased" words, we can easily detect correct values based on detection of these words. If not, we can detect all items of invoice, and we prepare machine learning model that classify which items they are by much sample invoice types and keywords/signs. If I can see your invoices samples, I can suggest more correct answer/offer for your project.

Dear client, Thanks a lot for taking your precious time to read my message. After browsing your job description, I am very interested in your project and I believe I’m qualified for the task. Regarding OCR, I have +10 years in this field and have worked on many successful image processing projects in the past. If you are looking for a highly skilled and rich experienced OCR expert with a deep knowledge, a professional attitude, excellent communication skills and the highest code quality, then I'm the person you are looking for. I'd like to talk more about this. If given a chance, I am highly confident in my ability to deliver the highest quality. I am confident that my involvement in your project will bring it to a successful launch, on time and within budget. I look forward to hearing from you. Kind Regards,

Hi, there! I am interested in your project. I have extensive experience in image processing and computer vision such as handwritten text recognition, ID card recognition, captcha recognition, ALPR(ANPR) and OCR using Python, OpenCV, keras, CNN and Tensorflow Faster R-CNN, Mask R-CNN, YOLO, SSD Inception and SSD Mobilenet. As a developer with rich experience, I am confident about your project. I will provide perfect results in a short time. If you want to work with me, please feel free to contact me. Thanks. PS: The method to recognize for handwritten text and standard text are different. If we recognize standard text, we can use OCR library such as Tesseract. In this case to recognize handwritten text, we must use train the text by using deep learning algorithm such as CNN, Keras, Tensorflow and etc.

I have extensive experience in the domains of Natural Language Processing, Recommender Systems,Machine learning,Data Minning, Deep Learning, Computer Vision, and Reinforcement Learning. While working in the domains mentioned above, I have worked used frameworks like TensorFlow, Scikit-learn, PyTorch,STP and UDP, Caffe2, OpenCV, Keras, matplotlib, seaborn, fastai, Pycharm,NILM algorithms etc) I am also very familiar with classification and clustering algorithms such as SVM,FHMM, Combinatorial Optimization, HART 85 Algorithm, Maximum Likelihood Estimation, KNN, LDA, Fuzzy classification I’m honest, creative and a unique person to carry out your tasks given to me. Leaving your project on our hand is a privileged; we'll make sure that the project is done in a perfect way and do our best until you satisfied. I would love to discuss this project with you and meet your exact requirements. I am confident I can provide you with top notch materials that will fit your needs. I'm proud my work on being professional, of the highest quality and always delivered on time. you can hire me directly through this link https://www.freelancer.com/u/olaideejiwole

hello we have read the description very well you ask 2 important questions and here is the answer first we can track the currency when the location varies as ( $ CAD , INR , etc ) second we can minimize the edits of the mistake by searching first of many types and styles if invoices then try to make an algorithm with many conditions to minimize the mistake as possible and we can extract the table of items by its columns and put it in a formatted sort hope we listen from you soon

Hi 1. We can make use of some template of invoice from same vendor and modle will learn total price section. 2. Again it has to be template , there should be some table. 3. Then you will be thinking why we need machine learning. ? Because it can pick hand writing better than existing OCR like with 98% accuracy. I do have experience of applying Inception, DenseNet, ResNet on different data sets. I have been working on Machine Learning algorithms like Regression, SVM, Decsion Trees and Deep Learning RNN, CNN and GANs Lets discuss

Hello, I am pleased to inform you that I have extensive experience of 6 years in the field of OCR and have achieved model with over 95% accuracy in recognition. As per your requirements if the invoices are typed or printed then the accuracy will go up much more and if they are handwritten then we would need to train and predict the model simultaneously. To answer your questions: 1) If the location of the total price varies with the number of items, then we can simple run an recognition model for all the values we get in the particular region and form a decision matrix that return the highest valued index of that matrix, thus giving us the total value. 2)We could look for generic terms such as 'Items' , Commodities' etc. to decide on the table and then work our way around it. But for me to answer this more precisely, I would prefer having a look at some sample invoices. Please consider my proposal for this job, as I am one of the best in my field with a lot a of experience. I would provide you with the most accurate prediction model you can get . Thank you.

First, we will need a labeled data set of invoices and the location of the prices. This can be done using Google Vision API that does OCR scanning and returns the text and its coordinates. The coordinates are a 2D vector, which can be fed into a neural network autoencoder. After the encoder, we can take the unique feature, the position of the table's header and its title, append that to the feature vector, and complete the decoder. This way, the model will learn not only what a title is and price is, it will learn to generate it. We don't actually need the generation part, we can throw that out once the model achieves small loss. Then, we can attach an output layer that returns approximate coordinates of where it thinks the prices are relative to the title. Then, we can search for all locations given out by the Google API for coordinates that match that. For the rest of the table, we can repeat this step with each column and each column's title. Also, there are OCR technologies that extract data from images. Why don't you use those? I am Computer Vision researcher at Cornell University and I am doing research in unsupervised representation learning. I am an expert in this field.