In this paper, we propose a new no-reference image quality assessment (NR-IQA). The method makes use of local binary patterns (LBP) to label local textures of an image. These labels form a LBP map that can be used to measure the characteristics of image textures (texture map). Then, we compute the histogram of the texture map and weight each LBP label according to its saliency, which is obtained with a visual attention computational model. The weighted histogram is used as input to a regression method that estimates the quality of the image. Experimental results show that the proposed method achieves competitive prediction accuracy and outperforms other state-of-the-art NR-IQA methods. At the same time, the method is simple and reliable, demanding few computational resources, such as memory and processing time.
In this paper we present a method of texture synthesis which removes the need for users to set, or even understand, parameters which have an impact on the synthesized output. We accomplish this by first classifying each input texture sample into one of three texture types: regular, irregular and stochastic. We found that textures within a class were synthesized well with similar parameters. If we know the input texture class, we can provide a good starting set of parameters for the synthesis algorithm. Instead of requiring a user to manually select a set of parameters, we simply ask that the user tell us whether the synthesized texture is satisfactory or not. If the output is not satisfactory, we adjust parameters and try again until the user is happy with the output. In this implementation we use the image quilting method in [1], a texture synthesis algorithm, as well as texture classification. With small adjustments our method can be applied to other texture synthesis methods.
In this paper, we propose a document image classification framework based on layout information. Our method does not use OCR; hence, it is completely language independent. Still we are able to exploit text data by extracting text regions with a novel MSER-based approach. Our MSER formulation provides great robustness against text distortions in comparison to the existing one. We introduce two types of novel image descriptors supplemented with Fisher vectors, based on Bernoulli mixture model. Classifiers, based on aforementioned descriptors, are assembled into meta-classification system that is able to classify document in complex cases when individual classifier accuracy is poor. Our meta-classification system demonstrates low processing time comparable to a single classifier. We show that our method outperforms the existing ones by the means of classification accuracy for a wide range of documents of both well-known and machine-generated document datasets.