Research On Quality Evaluation Of Sketch Artworks Based On Hierarchical Comparisons

In art schools,in order to improve panting skill of students,a teacher needs to compare and score artworks of students with reference images as well as evaluate the quality of each artwork itself.Due to the conflicting criterion and human influences,using analytical and computational capabilities of computers to improve the efficiency and automation of the evaluation flow has an important research value.Sketch itself has the characteristics of artistic expression and diversification of evaluation criteria,bringing the assessment of sketch quality a lot challenges.In re-sponse to these difficulties,this paper decomposes the quality assessment of sketch artworks into three levels to study the problem systematically,which include overall assessment,sub-region assessment and stroke assessment.The main work of this paper is as follows:1.In order to evaluate the overall quality of sketch artworks,this paper proposes a processing method based on shape description and modeling.The method first uses Gaussian derivatives to screen and cluster pixels on the lines to get key points.Then we extract multi-level features in the local areas of key points based on HOG and LBP descriptor and combine them to a new shape descriptor.After that,the reference works are matched with the key points of the artworks based on the new descriptor to obtain the differences.Finally,a regression tree model is established based on the collected labeled data sets and the differences,and the global quality of the imitated work is automatically evaluated through the model.2.For the analysis of selected area,we first use uniform sampling methods to get points on the boundary,then describe and match the sampling points using shape context feature to obtain the assignment,finally qualitative analysis of the selected area is made according to the distance characteristics between matching pairs.3.As for the analysis of stroke quality,this paper first proposes a stroke width estimation method.The method first calculates the width based on the artificially de-fined four directions.Then,the smallest width among the four directions is selected as the estimated width,and the direction is taken as the estimated normal direction of the pixel.After that,we use the estimated width and normal direction to detect abnormal areas.Three methods are tested on the collected sketch data sets.Experiments show that our processing method can obtain good results for the global evaluation of sketch works,and the algorithm for subdivided regions fulfill the qualitative analysis of the region to some extent.The line quality analysis method can also accurately analyze some of the abnormal areas of the sketch works such as abnormal area of thickness.