Optimization method and system for stone carving character recognition
A technology of text recognition and optimization method, which is applied in the field of optimization method and system of stone inscription text recognition, which can solve the problems of high recognition ability and fuzzy scenes that cannot be realized, and achieve the effect of improving accuracy and service ability of scenic spots
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0026] This embodiment discloses as figure 1 An optimization method for recognition of stone inscriptions is shown, comprising the following steps:
[0027] S1 obtains pictures of stone carvings in special environments, and uses convolutional neural networks to train models of stone carvings in special environments;
[0028] After N iterations of S2, the parameters trained in the special environment stone carving model are superimposed on the daily stone carving training process of each layer, and used as the offset value after each layer of convolution to enter the next layer of convolution, where N is a natural number;
[0029] S3 extracts features in the convolutional layer, uses feature value deconvolution to restore the original picture, and restores the features of the original picture;
[0030] S4 performs convolution processing on the restored image, obtains the features of the generated image, and then confronts the original image, and finally obtains an optimized mo...
Embodiment 2
[0036] This embodiment provides an optimization method for stone inscription recognition, referring to figure 2 As shown, in this embodiment, the convolutional neural network is used to train pictures of stone carvings in a special environment, and the results corresponding to each picture are used as labels for training the model. Special environment stones are first trained individually for N iterations, where N is an integer. The model will first obtain a part of the stone inscription content features, and remove irrelevant features.
[0037] Then, the parameters after the training of the special environment stone carving model are superimposed on the daily stone carving training process of each layer, as the offset value after each layer of convolution, and enter the next layer of convolution. After the features are extracted by the convolutional layer, the original image is restored by using the feature value deconvolution, and the loss function is used to continuously ...
Embodiment 3
[0041] This embodiment provides a specific application of an optimization method for stone inscription text recognition, training sample data, selecting different fonts in a daily sunny environment, 10 different scenic spots, and 50 photos from different angles for each scenic spot, a total of 500 photos. Select the same 10 scenic spots for special environment training pictures. In rainy days, cloudy days and other links, 50 pictures are taken from the same angle, and a total of 500 pictures are used for training.
[0042] Convolution parameters:
[0043] Input image size: 2560x1920 pixels
[0044] Convolution 1, Convolution 2: 5 convolutional layers, filter size 10x10, using maximum pooling, offset: 10 pixels
[0045] Convolution 3: 5 convolutional layers, filter size 15x15, max pooling, offset: 10 pixels
[0046] Deconvolution 1: 5 convolutional layers, filter size 10x10, offset: 10 pixels
[0047] Learning rate: 0.005, when the loss function value is lower than 0.01, the...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 

