Akifumi Makinouchi

Early ERP components differentially extract facial features: evidence for spatial frequency-and-contrast detectors.

It is generally accepted that the N170 component of an event-related potential (ERP) reflects the structural encoding of faces and is specialized for face processing. Recent neuroimaging and ERP studies have demonstrated that spatial frequency is a crucial factor for face recognition. To clarify which early ERP components reflect either coarse (low spatial frequency, LSF) or fine (high spatial frequency, HSF) processing of faces, we recorded ERPs induced by manipulated face stimuli. By filtering the original grayscale faces (broadband spatial frequency) spatially, we created LSF and HSF face stimuli. Next, we created physically equiluminant (PEL) face stimuli to eliminate the effects of lower order information, such as luminance and contrast. The P1 amplitude at the occipital region was augmented by LSF faces, while the N170 amplitude increased for HSF faces. The occipital P1 amplitude for PEL faces was relatively unaffected compared with that for PEL houses. In addition, the occipital N2 for PEL faces was spatiotemporally separable from N170 in a time-window between P1 and N170. These results indicate that P1 reflects coarse processing of faces, and that the face robustness further assures face-specific processing in the early component. Moreover, N2 reflects the early contrast processing of faces whereas N170 analyzes the fine facial features. Our findings suggest the presence of spatial frequency-and-contrast detectors for face processing.

Electrophysiological evidence for sequential discrimination of positive and negative facial expressions

OBJECTIVEnTo elucidate the differences in temporal processing between positive and negative facial expressions by using event-related potentials (ERPs) with spatially filtered images.nnnMETHODSnBased on the traits of parallel visual pathways, four types of facial expression images (happiness, fear, anger and neutral) with low, high and broadband spatial frequencies (LSF, HSF and BSF, respectively) were carefully created with the consideration of luminance, contrast and emotional intensity. These images were pseudo-randomly presented to 13 healthy subjects to record ERPs. Twenty recording electrodes were placed over the scalp according to the International 10-20 system. For emotion-relevant late negative components with latencies of 190-390 ms, the amplitude differences among the four facial expressions were analyzed for sequential 20-ms time windows by ANOVA.nnnRESULTSnThere were significant amplitude differences between positive and negative LSF facial expressions in the early time windows of 270-310 ms at the occipitotemporal region. Subsequently, the amplitudes among negative HSF facial expressions differed significantly in the later time windows of 330-390 ms.nnnCONCLUSIONSnDiscrimination between positive and negative facial expressions precedes discrimination among different negative expressions in a sequential manner based on parallel visual channels.nnnSIGNIFICANCEnERPs with spatially filtered images have provided the first evidence for sequential discrimination of positive and negative facial expressions.

Efficient evaluation of partially-dimensional range queries using adaptive r*-tree

This paper is about how to efficiently evaluate partially-dimensional range queries, which are often used in many actual applications. If the existing multidimensional indices are employed to evaluate partially-dimensional range queries, then a great deal of information that is irrelevant to the queries also has to be read from disk. A modification of R*-tree is described in this paper to ameliorate such a situation. Discussions and experiments indicate that the proposed modification can clearly improve the performance of partially-dimensional range queries, especially for large datasets.

Efficient evaluation of partially-dimensional range queries in large OLAP datasets

In light of the increasing requirement for processing multidimensional queries on OLAP (relational) data, the database community has focused on the queries (especially range queries) on the large OLAP datasets from the view of multidimensional data. It is well-known that multidimensional indices are helpful to improve the performance of such queries. However, we found that much information irrelevant to queries also has to be read from disk if the existing multidimensional indices are used with OLAP data, which greatly degrade the search performance. This problem comes from particularity on the actual queries exerted on OLAP data. That is, in many OLAP applications, the query conditions probably are only with partial dimensions (not all) of the whole index space. Such range queries are called partially-dimensional (PD) range queries in this study. Based on R*-tree, we propose a new index structure, called AR*-tree, to counter the actual queries on OLAP data. The results of both mathematical analysis and many experiments with different datasets indicate that the AR*-tree can clearly improve the performance of PD range queries, esp. for large OLAP datasets.

Using Filtering Algorithm for Partial Similarity Search on 3D Shape Retrieval System

Because 3D models are increasingly created and designed using computer graphics, computer vision, CAD medical imaging, and a variety of other applications, a large number of 3D models are being shared and offered on the Web. Large databases of 3D models, such as the Princeton Shape Benchmark Database [1], the 3D Cafe repository [2], and Aim@Shape network [3], are now publicly available. These datasets are made up of contributions from the CAD community, computer graphics artists, and the scientific visualization community. The problem of searching for a specific shape in a large database of 3D models is an important area of research. Text descriptors associated with 3D shapes can be used to drive the search process [4], as is the case for 2D images [5]. However, text descriptions may not be available, and furthermore may not apply for part-matching or similarity-based matching. Several content-based 3D shape retrieval algorithms have been proposed [6–8]. For the purpose of content-based 3D shape retrieval, various features of 3D shapes have been proposed [6–9]. However, these features are global features. In addition, it is difficult to effectively implement these features on relational databases because they include topological information. An efficient feature is proposed in Lu et al. [10] that can also be used in the partial similarity matching of shapes. However, they do not describe an efficient method to retrieve complex shapes by their partial similarity in Lu et al. [10] for a 3D shape database. In addition, the shock graph comparison-based retrieval method described in a previous paper [11] is based only on the topological information of the shape. A geometrical, partial similarity-based, and efficient method is needed to retrieve 3D shapes from a 3D shape database. In this chapter, we propose a novel filtering method to filter out shapes. The proposed method is based on geometrical information rather than on topological information alone. Shapes are removed from the candidate pool if the processing part of the key shape is not similar to any part of the potential candidate shape. We

A 3D object retrieval method using segment thickness histograms and the connection of segments

We introduce a novel 3D object retrieval method that is based on not only the topological information but also the partial geometry feature of 3D object. Conventional approaches for 3D object similarity search depend only on global geometry features or topological features. We use the thickness distribution along the segment of curve-skeleton as the partial geometry feature of 3D object and define the connection of the segments of curve-skeleton as the topological feature of 3D object. In order to retrieve 3D objects, we match 3D objects by their thickness distributions along segment on the curve-skeletons. Furthermore, we use the connection information of segments to improve the accuracy of partial similarity retrieval. The experimental evaluation shows that our approach yields meaningful results in the articulated object database.

Sorting AR*-tree: Further Improving the Performance of Partially-dimensional Range Queries

It is well known that multidimensional indices are helpful to improve the performance of range queries in multi-dimensional spaces. An n-dimensional index is often used for evaluating n-dimensional queries. However in many applications using range queries, the query dimensions of each range query are likely of only part (rather than all) of the index dimensions]. Such range queries are referred to as partially-dimensional (PD) range queries in our previous study (Feng and Makinouchi, 2006). That is, although the index is built in an n-dimensional space, the actual range queries may only use d dimensions of the n dimensional index space (d < n). If the existing multidimensional indices are employed to evaluate PD range queries, then a great deal of information that is irrelevant to the queries also has to be read from disk. In order to solve this problem, we proposed a modification of R*-tree, called Adaptive R*-tree (AR*-tree). This paper is about how to further improve the search performance of the AR*-tree for PD range queries by sorting the entries in AR*-tree nodes.

Multi-Part Registration of Regions using the Models derived from the Visible Human Male

Our research goal is to construct digital anatomy for individual patients. Since the shapes of organs are different patient by patient, registration of organs of different persons is an important research issue. In this paper, we present a new approach to multi-part affine registration which enables to register multiple regions of organs of different persons. We construct a standard anatomy of human heart and lung for the registration. It is derived from the Visible Human Male data set distributed by the United States National Institutes of Health. There are two types of models in the anatomy. They are voxel model and triangular mesh model. To construct the voxel model, we extracted seven regions in the heart, and seven regions in the lung from the Visible Human Male manually. We constructed the triangular mesh model from the voxel model.