Application Of Visual Perception Simulation Technology In Sketch Recognition

The human vision system has the ability to interpret complex scenes.When humans look at a scene,they are first attracted to one of the most compelling parts of the scene,this process called visually saliency.With the development of computer information science and artificial intelligence,machine vision and target significance detection play more and more important roles in daily life,industrial application and national defense and military affairs.The main contents of this paper include:Firstly,the input sketches are sorted by correcting the input sketches.The sketch is a fuzzy approximation to its intention,so the sketch is very fuzzy and random.It needs to extract the feature information in the sketch and correct the feature of the standard entity.The normative primitives are divided into the following: inflection points,straight lines,arcs,circles,and polygons.Using the characteristics of various primitives to correct it through the mathematical method.Corrected sketches can be divided into four categories: single closed,single open,multi-strokes closed and multi-strokes open.So,creating a sketch classifier according to the number of input sketches and whether it closed or not.The input sketch is classified for the comparison of the similarity of the sketch and the target.Then,an improved Itti algorithm based on corner weight is proposed.The traditional Itti significance model can get the salient regions of the image,but the final total saliency map is obtained by the linear combination of the salient graphs,while the salient regions are obtained is non-unique,and the weights of different salient regions are not taken into account.Inhibition and the winner to take the whole mechanism to focus on the regions transfer.In computer vision,corners are called feature points,which are important features of images.Therefore,this paper adds the corner features to Itti significant model,and get the most significant regions of the image directly.Lastly,compare the similarity between the input sketch and the salient region of the target image after the correction.The feature extracted in this paper is saved in the form of vector points,then calculate the Hu moments of the two sets of eigenvector points,and judge whether the two pictures are similar by the Hu moments.