Wavelet Image Compression Algorithm Combined with Neural Network for Feature Points in the Visual Measurement Area

In this paper, the feature point region compression in the visual measurement is encoded. First, the wavelet decomposition is used to obtain the high-frequency decomposition sub-image and the low-frequency decomposition sub-image. Second, the low-frequency signal is vectorized based on the improved neural network to form a quantized image. Then, the obtained high-frequency and low-frequency components respectively perform feature point coding point detection based on region statistics to identify the feature point region, and the feature point region and the background region are respectively processed. The feature point region is losslessly compressed by integer wavelet transform, and the background region adopts discrete cosine. The lossy compression is transformed, and finally the high and low frequency images are reconstructed to obtain the final compressed image. Besides, in order to display the characteristics of the algorithm more intuitively, the compressed image is compared with TIFF compression, JPEG compression and literature 3 compression. The experimental results show that the compressed image obtained by the algorithm is high fidelity under the premise of guaranteeing the compression ratio, that is, the feature point region. The information is completely saved. It is suitable for the study of lossless compression of feature points in visual measurement.