Abstract: For the huge information and poor adaptability of segmentation effect in the colorized images，this article proposes a method to improve the accuracy of extracting and assorting the image semantics.First control the segmentation precision of the image semantics region and extract the image texture characteristics.Then test the samples’ classification through the improved probabilistic neural network model.The experiment shows that the precision of assorting the colorized images semantics region through the improved probabilistic neural network is 90% while it’s 70% before.The new method gets the better classification effect.