Video Aesthetic Quality Assessment For Multi-type Features

Medical Image Segmentation have been widely used in multiple industrial application,such as : clinical diagnosis,medical research,surgical planning,pathological analysis,image information processing and computer-assisted surgery.In the field of medical research,the most important usage of Medical Image Segmentation is to extract target area,thus assisting doctors to analyze medical images based on extraction results.Furthermore,the ratio of medical images from different regions,the quantitative measurement of organs after anatomy,the identification of cells and the statistics of numbers and the division of organs will help doctors draw up treatment and surgical plans.Meanwhile,Medical Image Segmentation compresses and transmits data remotely for hospitals.Moreover,the results of Medical Image Segmentation can also be used to create a medical image database,which greatly reduces the resources required for medical data access and search.However,medical images suffer from heavy noise and poor quality,which are caused by the characteristics of the medical images and the cameras.Firstly,the characteristics of medical images includes high similarity between the tissue structures and the blurry boundaries of the images,which results in low contrast between the lesions and the background.Secondly,the size and shape of the lesion area various from different individuals.In this case,it is harder to identify the lesion area for certain diseases.Thirdly,cameras can also induce noise deviated from the natural image(e.g.,quantum noise,particle noise,current noise,etc.),which results in a large grayscale difference between the structural edges and the noise in the medical image and the background.Meanwhile,there exists visible noises in medical images such as stripe,fine grains,spots,snowflake shape and Black or white structure.Therefore,the quality of medical images is poor.To reduce the high noise and refine the low quality problems in medical images,the contribution of this paper are listed as below:(1)This paper proposes a novel network model for processing high-noise and low-quality medical image segmentation tasks.The model contains a multi-scale feature extraction module and an advanced feature auxiliary extraction module.Besides,this paper optimized the noise suppression and loss function of the preprocessing,and merged the traditional noise reduction methods with deep learning.(2)This paper proposes an attention-based network model for processing high-noise and low-quality medical image segmentation tasks.The model includes a self-attention module and a cross-attention module to capture long-distance dependence to reduce the adverse effects of high noise and low quality.(3)According to the proposed network model,this paper has done a series of experiments on the skin cancer,lung cancer,and liver data sets.The results prove the superior effectiveness of segmentation over compared approaches and the effective contribution of each component in proposed network model.