| The jigsaw puzzle algorithm is a computer algorithm to reassemble a set of randomly disrupted fragments into a complete image,which is widely used in digital image restoration,computer vision,and intelligent games.Square fragment restoration is a hot research topic in jigsaw puzzle algorithms.The difficulty lies in finding the neighboring blocks accurately from many similarly shaped fragments.The traditional square puzzle algorithm relies on the similarity calculation between fragment blocks and lacks the understanding of the position relationship of neighboring blocks,resulting in "garbled" puzzle results.Meanwhile,the traditional algorithm has problems such as large search space in the image restoration task,which makes it difficult to meet the real-time requirements in the real scene.In response to the above problems,this paper focuses on the jigsaw reduction method of square digital images,and the specific work is shown below:(1)To improve the accuracy of the similarity metric results and the robustness of the restoration results,this paper proposes an optimization method for jigsaw puzzle restoration based on the DPJaccard-SSD metric.First,since there is no open-source large fragment dataset in the field of jigsaw puzzles,this paper adopts PASCAL VOC 2007 as the initial dataset and performs classification,image scaling and uniform sampling operations to construct the corresponding fragment dataset.Secondly,the traditional Jaccard metric is improved from both array partitioning and feature fusion,and a DPJaccard-SSD metric is proposed to measure the similarity between fragment blocks,adding its boundary information in the process of adjacency finding,which improves the mean value of the optimal metric ratio by 6.25%on the test dataset.Finally,the Kruskal-based splicing strategy prioritizes the fragment blocks that satisfy the "Best Buddies"relationship for the restoration operation.Compared with the Jaccard metric-based jigsaw puzzle restoration algorithm,the restoration results obtained by this method improve by up to 6.12%in the Direct Comparison metric and up to 6.42%in the Neighbor Comparison metric,and the experimental results further validate the effectiveness of the method.(2)In order to preferentially select adjacent blocks with higher confidence levels for the puzzle restoration task,this paper proposes an optimization method for jigsaw puzzle restoration based on JSSD metric and data filtering.The method firstly uses the JSSD metric based on waterfall fusion to measure the similarity between fragment pairs,locates the best neighboring blocks through layerby-layer data filtering,and constructs the initial relationship matrix based on the similarity metric.Secondly,the data filtering module based on the position constraint is added to the jigsaw puzzle restoration process,and the data in the relationship matrix is filtered by using the two criteria of "Best Buddies"and the convergence theorem to improve the accuracy of the square puzzle restoration.Finally,based on the greedy idea,the fragmented tiles are gradually stitched together,and the fragments are restored by combining pruning and filling techniques.This paper conducts experiments on 12 sets of test images.The experimental results show that the jigsaw restoration method based on JSSD metrics and data filtering is effective in the restoration ta sk,and its average accuracy rate can reach 98.08%.(3)An intelligent jigsaw puzzle restoration system has been designed based on an improved jigsaw puzzle optimization method.With a simple and intuitive user interface and interactive operation,this system can quickly help users construct fragmented data sets and digital image restoration. |
