In this letter, we introduce a novel patch sampling strategy for the task of image classification, which is fundamentally different from current patch sampling strategies. A top-down guidance learned from training images is used to guide patch sampling towards informative regions. Experiment results show that this approach achieved noticeable improvement over baseline patch sampling strategies for the classification of both object categories and scene categories.

We propose a new multi-objective genetic programming (MOGP) for automatic construction of image feature extraction programs (FEPs). The proposed method was originated from a well known multi-objective evolutionary algorithm (MOEA), i.e., NSGA-II. The key differences are that redundancy-regulation mechanisms are applied in three main processes of the MOGP, i.e., population truncation, sampling, and offspring generation, to improve population diversity as well as convergence rate. Experimental results indicate that the proposed MOGP-based FEP construction system outperforms the two conventional MOEAs (i.e., NSGA-II and SPEA2) for a test problem. Moreover, we compared the programs constructed by the proposed MOGP with four human-designed object recognition programs. The results show that the constructed programs are better than two human-designed methods and are comparable with the other two human-designed methods for the test problem.

Bag of visual words is a promising approach to object categorization. However, in this framework, ambiguity exists in patch encoding by visual words, due to information loss caused by vector quantization. In this paper, we propose to incorporate patch-level contextual information into bag of visual words for reducing the ambiguity mentioned above. To achieve this goal, we construct a hierarchical codebook in which visual words in the upper hierarchy contain contextual information of visual words in the lower hierarchy. In the proposed method, from each sample point we extract patches of different scales, all of which are described by the SIFT descriptor. Then, we build the hierarchical codebook in which visual words created from coarse scale patches are put in the upper hierarchy, while visual words created from fine scale patches are put in the lower hierarchy. At the same time, by employing the corresponding relationship among these extracted patches, visual words in different hierarchies are associated with each other. After that, we design a method to assign patch pairs, whose patches are extracted from the same sample point, to the constructed codebook. Furthermore, to utilize image information effectively, we implement the proposed method based on two sets of features which are extracted through different sampling strategies and fuse them using a probabilistic approach. Finally, we evaluate the proposed method on dataset Caltech 101 and dataset Caltech 256. Experimental results demonstrate the effectiveness of the proposed method.

The extened Kalman filter (EKF) and unscented Kalman filter (UKF) have been successively applied in particle filter framework to generate proposal distributions, and shown significantly improving performance of the generic particle filter that uses transition prior, i.e., the system state transition prior distribution, as the proposal distribution. In this paper we propose to use the Gauss-Newton EKF/UKF to replace EKF/UKF for generating proposal distribution in a particle filter. The Gauss-Newton EKF/UKF that uses iterated measurement update can approximate the optimal proposal distribution more closer than EKF/UKF, especially in the case of significant nonlinearity in the measurement function. As a result, the Gauss-Newton EKF/UKF is able to generate and propagate the proposal distribution for each particle much better than EKF/UKF, thus further improving the performance of state estimation. Simulation results for a nonlinear/non-Gaussian time-series demonstrate the superior estimation accuracy of our method compared with state-of-the-art filters.

In this paper, we evaluate a prediction method of regions including dislocation clusters which are crystallographic defects in a photoluminescence (PL) image of multicrystalline silicon wafers. We applied a method of a transfer learning of the convolutional neural network to solve this task. For an input of a sub-region image of a whole PL image, the network outputs the dislocation cluster regions are included in the upper wafer image or not. A network learned using image in lower wafers of the bottom of dislocation clusters as positive examples. We experimented under three conditions as negative examples; image of some depth wafer, randomly selected images, and both images. We examined performances of accuracies and Youden's J statistics under 2 cases; predictions of occurrences of dislocation clusters at 10 upper wafer or 20 upper wafer. Results present that values of accuracies and values of Youden's J are not so high, but they are higher results than ones of bag of features (visual words) method. For our purpose to find occurrences dislocation clusters in upper wafers from the input wafer, we obtained results that randomly select condition as negative examples is appropriate for 10 upper wafers prediction, since its results are better than other negative examples conditions, consistently.

We studied the detection of the incongruence between the two eyes' retinal images from occlusion perception. We previously analyzed the evasion action caused by occlusion by using green-red equiluminance, which is processed by parvocellular cells. Here we analyzed this action by using yellow-blue equiluminance, which is said to be treated by koniocellular cells and parvocellular cells. We observed that there were the cases in which the subject could perceive incongruence by the occlusion and other cases in which the subject could not perceive it. Significant differences were not seen in all conditions. Because a difference was seen in an evasion action at the time of the rim occlusion gaze when we compare the result for the yellow-blue equiluminance with the green-red equiluminance, it is suggested that the response for each equiluminance is different. We were able to clarify the characteristic difference between parvocellular cells and koniocellular cells from an occlusion experiment.

Stereoscopic vision technology is applied in a wide range of fields, from 3D movies to medical care. Stereoscopic vision makes it possible to observe images in parallax between both eyes. However, parallax images cannot be used all the time due to a situation called “occlusion”, in which an object is hidden in the depths by another object. In this case, different images are projected on the right and left retina. Here, we propose a psychology experiment to elucidate the function of parvocellular cells in the LGN of the visual cortex of the brain using occlusion perception. As a new psychology experiment to clarify whether parvocellular cells in the LGN of the visual cortex, said to process chromatic and luminance information, can detect a disagreement between the retinal images produced by each eye, we measured convergence eye movement when gazing at the rim of a column under occlusion using an equiluminance random dot pattern. When eye movement prevented the disagreement of the retinal images caused by occlusion, we thought that convergence eye movement to move both eyes in front of the rim or divergence eye movement to move them behind the rim would occur. In other words, we thought that we could clarify whether there was parvocellular system process agreement or disagreement between the right and left retinal images under equiluminance. Therefore, we examined whether a system to detect disagreement between the retinal images exists in the brain when gazing at the rim of a column onto which an equiluminance random dot texture was mapped. Results suggested that the mechanism to avoid disagreement between the retinal images of the eyes caused by occlusion occurs in the parvocellular cells, which mainly process color information, as well as in the magnocellular cells, which process binocular disparity.

We propose a learning strategy for acceleration in learning speed of genetic programming (GP), named hierarchical structure GP (HSGP). The HSGP exploits multiple learning nodes (LNs) which are connected in a hierarchical structure, e.g., a binary tree. Each LN runs conventional evolutionary process to evolve its own population, and sends the evolved population into the connected higher-level LN. The lower-level LN evolves the population with a smaller subset of training data. The higher-level LN then integrates the evolved population from the connected lower-level LNs together, and evolves the integrated population further by using a larger subset of training data. In HSGP, evolutionary processes are sequentially executed from the bottom-level LNs to the top-level LN which evolves with the entire training data. In the experiments, we adopt conventional GPs and the HSGPs to evolve image recognition programs for given training images. The results show that the use of hierarchical structure learning can significantly improve learning speed of GPs. To achieve the same performance, the HSGPs need only 30-40% of the computation cost needed by conventional GPs.

We investigated the function in the Lateral Geniculate Nucleus of avoidance behavior due to the inconsistency between binocular retinal images due to blue from vergence eye movement based on avoidance behavior caused by the inconsistency of binocular retinal images when watching the rim of a blue-yellow equiluminance column.