Media Content Analysis Based On Exemplar Information Transfer

Multimedia information processing is the key technology of artificial intelligence,and plays an important role in various fields.When analyzing media content,it is often necessary to solve the semantic gap.A feasible idea is that:based on the establishment of geometric correspondences between media data,the key information on the exemplar data is transferred to the target data.Based on this idea,we carried out the research of 3D shape analysis and natural image processing,and achieved the application of anatomical landmark detection on 3D human shapes and image contrast enhancement.The outline is as follows:In the second chapter,we propose a method for anatomical landmark detection on 3D human shapes.Due to the progress of 3D shape acquisition technology,the geometry of the human body surface can be accurately reconstructed.The location of anatomically landmark on the human body surface has great importance in many industrial applications.Although these tasks can be accomplished by manual annotating,it is time-consuming and labor-consuming.Therefore,the automatic detection method is supposed to solve the problem where the geometric feature of the target landmark is not obvious.For this purpose,based on the core idea that establishes the correspondence between the exemplar shape and the target shape,our method transfers the position information of the annotated landmarks on the exemplar shape to the target shape.The entire detection framework is divided into exemplar selection,shape correspondence calculation,fine search and late fusion.The gradual improvement of the anatomical landmark detection accuracy has been achieved.Multiple kinds of shape features are used in different parts of our method.In the experiments based on the SHREC14 dataset,we validate the effectiveness of each part in the proposed method,and the comparison experiments with other related methods showed the state-of-the-art performance of our method.In the third chapter,we propose an image contrast enhancement method based on high quality exemplars.The popularization of smart phones makes it easier to take pictures.Due to the lack of photographing skills and terrible lighting conditions,the photos usually have poor aesthetic quality.In order to improve the aesthetic quality of these images,editing software is an acceptable choice.However,it usually requires many kinds of human interactions and sufficient manual editing skills,which is inefficient when dealing with a large number of images.We propose an image contrast enhancement model by establishing the feature matching between the exemplar image and the target image,which can automatically improve the aesthetic quality of the target image.Of note,the aesthetic quality of the exemplar image plays a key role in this model.Therefore,we use an aesthetic evaluation network to obtain the aesthetic score,where only high quality images are selected as the exemplar images.Visual and quantitative experiments validate our improved model.