Project description
Programmer services are required. The programmer is expected to interact (discussion of algorithms, explanation of code, joint ideas for improving the program, etc.). P.). The choice of the implementation language is discussed and not very important. We'll negotiate a price)
Setting the task:
A graphic file is supplied to the system (in my case it will be a photo of the car, the purpose is to recognize its number in the format “xx000x”) – all photos will be prepared by me.
At the output, the system outputs a text file in which the recognized characters are indicated in the ordinary text in the order and with the line transfer as they were specified in the original file (in my case without moving lines).
The system is based on the use of a multilayer perceptron, as well as on the algorithm for isolating each individual character of the source file and recognizing it using a neural network (N neural networks / 1 neural network with N outputs). Unfortunately, the use of special libraries is not allowed, you will have to prescribe everything in detail, cycles in a cycle, various things inherent in the NS - activation functions, calculations, passages, etc.
The system must be able to recognize at least 20 different symbols (digits, letters of the Russian alphabet, register - a single one).
Accuracy of recognition is not lower than 90%.
Recognition of each individual symbol should be implemented by the neural network. To search for a photo with a machine plate with a number, you need to use the Viola-Jones method (after detecting a plate with a number, it must be divided into symbols, and each of them is recognized by a neural network).
To train the NS I will prepare the input data (this will be symbols on a white background, the number of image options for each character can be discussed, about 3-5 I think will be enough).
Setting the task:
A graphic file is supplied to the system (in my case it will be a photo of the car, the purpose is to recognize its number in the format “xx000x”) – all photos will be prepared by me.
At the output, the system outputs a text file in which the recognized characters are indicated in the ordinary text in the order and with the line transfer as they were specified in the original file (in my case without moving lines).
The system is based on the use of a multilayer perceptron, as well as on the algorithm for isolating each individual character of the source file and recognizing it using a neural network (N neural networks / 1 neural network with N outputs). Unfortunately, the use of special libraries is not allowed, you will have to prescribe everything in detail, cycles in a cycle, various things inherent in the NS - activation functions, calculations, passages, etc.
The system must be able to recognize at least 20 different symbols (digits, letters of the Russian alphabet, register - a single one).
Accuracy of recognition is not lower than 90%.
Recognition of each individual symbol should be implemented by the neural network. To search for a photo with a machine plate with a number, you need to use the Viola-Jones method (after detecting a plate with a number, it must be divided into symbols, and each of them is recognized by a neural network).
To train the NS I will prepare the input data (this will be symbols on a white background, the number of image options for each character can be discussed, about 3-5 I think will be enough).
Software development Cash, Bank transfer
