| Summary: | With the rapid development of computer vision and artificial intelligence, image conversion technology has been widely applied in the fields of digital media and art. Based on the Generative adversarial network, this paper proposes a painting image conversion algorithm for multi class scenes. It performs deep feature extraction in the residual module, divides the encoding and decoding parts of the generator into functional parts, and designs them separately. The mixed attention module is inserted into the decoder and encoder to preserve the texture details of the image. The deep network interpolation is incorporated to achieve smooth and continuous conversion of the painting image. The experiment showed that the loss value of the research method decreases to 0.008 after 400 iterations during the loss value testing. Its maximum peak signal-to-noise ratio is 34.9dB when the bit rate increases to 1000kb/s. In the SAR image conversion dataset, the F1 value increases to 97.4 after 200 iterations. The pixel loss when it reaches 100% conversion in outdoor images is 5.38k. The data indicates that the research method has good performance in painting image conversion and can provide effective technical references for image conversion.
|
|---|