Hao-Chiang Shao

Optimal Multiresolution Blending of Confocal Microscope Images

Typical mosaicing schemes assume that to-be-combined images are equally informative; thus, the images are processed in a similar manner. However, the new imaging technique for confocal fluorescence images has revealed a problem when two asymmetrically informative biological images are stitched during microscope image mosaicing. The latter process is widely used in biological studies to generate a higher resolution image by combining multiple images taken at different times and angles. To resolve the earlier problem, we propose a multiresolution optimization approach that evaluates the blending coefficients based on the relative importance of the overlapping regions of the to-be-combined image pair. The blending coefficients are the optimal solution obtained by a quadratic programming algorithm with constraints that are enforced by the biological requirements. We demonstrate the efficacy of the proposed approach on several confocal microscope fluorescence images and compare the results with those derived by other methods.

Colored multi-neuron image processing for segmenting and tracing neural circuits

Recently developed were the Brainbow and Flybow techniques that can image and visualize a large number of neurons simultaneously; however, scientists still lack adequate tools to process this kind of colored multi-neuron image volumes. Due to dozens of colorized neuron fibers spreading densely in a very intricate structure, it is difficult to trace them by existing algorithms designed for single-neuron images. We proposed a framework to formulate and solve this issue, and the experimental results show that our method can successfully extract independent neurons from Flybow images. Consequently, the proposed procedure contributes to neuroscience by increasing the efficiency of collecting neuron information from Flybow images.

Retention of Features on a Mapped Drosophila Brain Surface Using a Bézier-Tube-Based Surface Model Averaging Technique

Model averaging is a widely used technique in biomedical applications. Two established model averaging methods, iterative shape averaging (ISA) method and virtual insect brain (VIB) method, have been applied to several organisms to generate average representations of their brain surfaces. However, without sufficient samples, some features of the average Drosophila brain surface obtained using the above methods may disappear or become distorted. To overcome this problem, we propose a Bézier-tube-based surface model averaging strategy. The proposed method first compensates for disparities in position, orientation, and dimension of input surfaces, and then evaluates the average surface by performing shape-based interpolation. Structural features with larger individual disparities are simplified with half-ellipse-shaped Bézier tubes, and are unified according to these tubes to avoid distortion during the averaging process. Experimental results show that the average model yielded by our method could preserve fine features and avoid structural distortions even if only a limit amount of input samples are used. Finally, we qualitatively compare our results with those obtained by ISA and VIB methods by measuring the surface-to-surface distances between input surfaces and the averaged ones. The comparisons show that the proposed method could generate a more representative average surface than both ISA and VIB methods.

Developing a Stereotypical Drosophila Brain Atlas

Brain research requires a standardized brain atlas to describe both the variance and invariance in brain anatomy and neuron connectivity. In this study, we propose a system to construct a standardized 3D Drosophila brain atlas by integrating labeled images from different preparations. The 3D fly brain atlas consists of standardized anatomical global and local reference models, e.g., the inner and external brain surfaces and the mushroom body. The averaged global and local reference models are generated by the model averaging procedure, and then the standard Drosophila brain atlas can be compiled by transferring the averaged neuropil models into the averaged brain surface models. The main contribution and novelty of our study is to determine the average 3D brain shape based on the isosurface suggested by the zero-crossings of a 3D accumulative signed distance map. Consequently, in contrast with previous approaches that also aim to construct a stereotypical brain model based on the probability map and a user-specified probability threshold, our method is more robust and thus capable to yield more objective and accurate results. Moreover, the obtained 3D average shape is useful for defining brain coordinate systems and will be able to provide boundary conditions for volume registration methods in the future. This method is distinguishable from those focusing on 2D + Z image volumes because its pipeline is designed to process 3D mesh surface models of Drosophila brains.

3D thin-plate spline registration for Drosophila brain surface model

With the progress of model averaging algorithms, scientists in the field of brain research have an increasing demand for methods capable to register and warp source data to the pre-registered standard atlas. We here propose a thin-plate spline (TPS) based surface registration method to facilitate the registration and warping process of Drosophila brain data. Our contributions are twofold. First, the proposed method performs TPS-based registration in the parameterization domain, and hence it no longer needs a rigid transformation to globally align and scale the input models. Second, the obtained well-registered surface model can act as boundary constraints for further volumetric registration schemes. Experiments show that the proposed method is effective. For models with a 750-voxel-long bounding box diagonal, the average surface-to-surface distance is reduced to about 0.1-voxel-long after registration.

A backward wavelet remesher for level of detail control and scalable coding

Multi-resolution and wavelet analysis have generated considerable interest in the field of mesh surface representation. In this paper, we propose a backward, coarse-to-fine framework that derives a semi-regular approximation of an original mesh, and demonstrate its effectiveness on level-of-detail and scalable coding applications. The framework is flexible and simple because the position of a new vertex at a finer resolution can be derived in a closed form, based on the affine combination of a subdivision scheme, the original mesh, and “new” information about the wavelet coefficients. We report the results of experiments on both applications; and also compare the scalable coding results with those of other methods.

Light field upsampling by joint bilateral filtering on epipolar plane images

Due to the trade-off between spatial and angular resolution, the effective spatial resolution of a light field image is usually less than one percent of the number of pixels on the photo sensor. In this paper, we propose a prototype algorithm to upsample a light field image. Because the boundary edges of 3D objects would result in lines on epipolar plane images (EPIs), the main idea of our method is to preserve these line structures while upsampling so that the image enlargement can still have sharp boundary edges. The kernel of the proposed algorithm is an iterative joint-bilateral filtering process. Experiments show that the upsampled image derived by our method is still refocusable and has better visual quality than those derived by other methods. Finally, the main contribution of this method is that it decomposes a 4D light field space L(u, v, s, t) upsampling problem into a series of 1D, parameter-free upsampling subproblems that can be solved fast in u-s and v-t EPI domains.

A surface-constrained volumetric alignment method for image atlasing

In this paper, we propose a prototype system capable of incorporating 3D shape information with conventional TPS-based (thin-plate-spline) volumetric registration method for image atlasing. Our method consists of two phases. The former phase registers and warps the 3D mesh surface models describing the tissue shape boundary of the input image volumes, and the latter aims to align the input image volumes with the aid of the boundary constraints suggested by the former. The proposed volumetric registration method is driven and constrained by the pre-registered 3D mesh surface model. Experiments show that using our framework for volumetric image registration and warping obtains a performance comparable to or better than a well-known benchmark method.