Research On Orthophoto Mosaicking And Quality Evaluation Methods Based On Semantic Segmentation

Digital orthophoto map(DOM)plays an important role in many geographic information platforms as indispensable data source,as one of the most significant 4D products in the field of phototopography,which can provide up-to-date and basic geographic information data for the construction of digital cities,monitoring of geographical conditions and earthquake relief work,etc.In general,these applications need a orthophoto map which can cover a large geographical area,but limited by the size of the sensor,a single orthophoto map is usually made by a series of images which are corrected by orthopedics.Generally speaking,the determination of the seamline is one of the key steps in the image mosaic.To determine whether mosaic results are consistent with people's subjective feelings,mosaic image quality evaluation is introduced as the main mean.The quality evaluation methods' relative deficiency also limits the development of mosaic technology to some extent,so it has practical significance to evaluate the mosaic results of digital orthophoto maps.An optimal seamline should avoid to go through most of the visual objects to maintain the integrity of the terrain,and avoid to go through most of the overlap region which has low image similarity and big dislocation of the terrain,thus minimizing the discontinuity along the seamline due to the geometric and radiation difference in the final mosaic results.The existing DOM's mosaic algorithms mainly create cost map through the input images,and then further detect the optimal seamline based on the cost map through different optimization algorithms.However,the color,texture,edge detection and other characteristics can't completely independently describe the overlapping image differences,and it is difficult to find the best combination.Unlike these methods,this paper proposes a mosaic algorithm for DOM based on semantic segmentation.At first,we use OESM model data as ancillary data to train the U-net model,and then we use the trained U-net model to generate a cost map for overlapping area images,refining to generate the initial seamline network,in the end we use the A*search algorithm for the optimal seamline detection.Because the existing mosaic image quality evaluation methods generally focus on one aspect of the quality evaluation:the radiation consistency or the geometric consistency,lacking of the comprehensive quality evaluation of the image mosaic,this paper presents a quantitative evaluation method,which is based on the buffer images and includes both the radiometric consistency and geometric consistency indexes.Firstly,the buffer zones are built around the seamline,and the buffer images are obtained from the left and right images corresponding to the seamline.Then based on this,the color difference and geometric dislocation on both sides of the seamline are quantitatively evaluated.Among this,the radiometric consistency is mainly evaluated by the indexes of the gray contrast coefficient and the uniform intensity-distribution,and the geometric consistency is mainly evaluated by using the offset distance of the SURF(speeded-up robust features)feature matching points.To evaluate the performance of the two proposed approaches,corresponding comparative experiments are conducted respectively.For DOM mosaic,the EEOSD method(Edge-Enhanced Optimal Seamline Detection,EEOSD)and OrthoVista method are selected to compare with the method in this paper,and two overlapping DOMs from the katovitz region of Poland are used for the experiment.The qualitative and quantitative comparison and analysis are performed respectively for the experimental results from the respect of the artificial quality evaluation and the respect of the DOM mosaic quality evaluation presented in this paper.Experimental results show that our mosaic algorithm can effectively make the seamline avoid to go through obvious ground objects,and keep the consistency of radiation and geometric around the seamline at the same time,and our method has higher computing efficiency.For DOM mosaic quality evaluation,selecting the urban orthophotos and mountain orthophotos as data respectively,we perform the quality evaluation on the mosaic results of three kinds of software,such as ArcGIS,Ortho Vista and DPGrid.and compared with the results of visual quality evaluation,we draw a consistent conclusion,thus verified the effectiveness of the quality evaluation method for DOM mosaic proposed in this paper.