When you move into a new home, you need to arrange the living room furniture. But furniture placement is challenging because it requires jointly optimizing a variety of functional and visual criteria. So, many researchers have developed computer-aided systems that help people, who have no prior knowledge on interior design, to arrange the furniture layout. Merrell et al. developed such a system based on a set of interior design guidelines. This system generates layout suggestions by using some cost function. But this system layouts for a single room. In this paper, we extend the Merrell’s method to layouts for multiple rooms by adding a new translation rule and cost functions. We test our system whether we can suggest layouts for multiple rooms.
In JPEG encoded images, the mosquito noise which is one of visual distortion is occurred around edge region in the case of lower bit rate. It is difficult to estimate the mosquito noise from any encoded image, because the subjective assessment for mosquito noise is dependent with a given image contents. This paper presents a development of quantitative evaluation measure for deterioration of image quality by mosquito noise using cepstrum coefficients. It is seen in our experimental results that mosquito noise can estimate by focusing on higher cepstrum coefficients in each encoded block.
In this talk, we make reference to practical cases of study from visual information processing in order to clearly explain the academic significance of encompassing the pipeline from visual data production to consumption as well as getting conscious about the research spiral with a period of two decades.
We conducted experiments to investigate the effect of display positions and orientations on visibility of in-vehicle display systems. In experiments, we installed display devices (tablet PCs) onto nine typical visual target positions on a vehicle (e.g., meter panel, left, right and inner mirrors, etc.) and evaluated visibilities of those devices under three illumination conditions (night, twilight, daylight). We employed RDK (Random Dot Kinematogram) as visual stimuli and determined PSE (Point of Subjective Equality) for displayed stimuli sizes between two devices (one is on the HUD (Head Up Display) position as a reference stimuli and another one is on one of nine devices including the HUD position itself.) through staircase procedures. Experimental results show that PSE for the left-mirror position device is significantly higher than PSE for other devices. Also, comparing PSE at door mirror positions with ones at positions where CMS (camera monitoring system) displays will be installed (just inside doors), we found visibility at latter position is better for the left side positions.
Prediction error expansion (PEE) is the method of reversible data hiding which makes a prediction error histo-gram, and expands prediction errors nearby zero for embedding information bits. Therefore, capacity of the embeddable bits depends on the kurtosis of the histogram. In this paper, we propose a watermarking method using a prediction based on the local directional interpolation (LDI). In the proposed method, prediction accuracy is improved, and can yield a better predic-tion error histogram for embedding. Experimental results show that the proposed method increases the maximum capacity of embedding information and improves the image quality in some images.
It was reported that Random noise is effective to improve the surface texture. When the impression of surface texture changes by adding random noise, we need to estimate the appropriate noise intensity for the improvement of surface texture. In this study, we have investigated the relationship between fractal dimension of the noise added image and the subjective evaluation of the image. It is seen in our experimental results using some texture images that the fractal dimension increases steeply when the noise intensity is small, and the amount of change of the fractal dimension is small or constant according to the noise intensity is large. It is seen in our experiments that the appropriate noise intensity for the improvement of surface texture can be obtained using difference of fractal dimension, and the proposed noise estimation measure can be applied to small region with 16×16 pixels in natural images.
The use of compressive sensing (CS) in applications with rapid spatial phase variations is difficult, since not only the magnitude but also phase regularization is required in the CS framework. We have proposed a novel image reconstruction scheme for 2D MR phase varied images in which phase regularizer is not required in the CS reconstruction scheme. In that method, to improve the incoherence between the sampling matrix and the basis of the sparsifying transform, multi-scale eFREBAS transform domain thresholding were used. In this article, we examined the effectiveness of multi-scale eFREBAS based CS to the 3D phase varied images. It was shown that higher quality of images were obtained compared to 2D single-slice CS.
An automated blood vessel extraction using high-order local autocorrelation (HLAC) on retinal images is presented. The blood vessels were enhanced by black-hat transformation as a pre-processing. One hundred five HLAC features were calculated in a black-hat filtered image, and they were inputted into a first layer of an artificial neural network (ANN). The output, a green component of the color retinal image, output values of three kinds of filter were then inputted into a second layer of ANN. Using DRIVE database, the area under the curve (AUC) based on ROC analysis was 0.960 as a result of our study. The result can be used for the blood vessels extraction.
When computer Shogi and Go research started they said computer never won against human top professional players. But now computer has surpassed human in Shogi and Go. I think Shogi and Go are good examples for artificial intelligence research to consider what happen when computer surpass human.