In this paper, we propose a deep hierarchical attention context model for lossless attribute compression of point clouds, leveraging a multi-resolution spatial structure and residual learning. A simple and effective Level of Detail (LoD) structure is introduced to yield a coarse-to-fine representation. To enhance efficiency, points within the same refinement level are encoded in parallel, sharing a common context point group. By hierarchically aggregating information from neighboring points, our attention model learns contextual dependencies across varying scales and densities, enabling comprehensive feature extraction. We also adopt normalization for position coordinates and attributes to achieve scale-invariant compression. Additionally, we segment the point cloud into multiple slices to facilitate parallel processing, further optimizing time complexity. Experimental results demonstrate that the proposed method offers better coding performance than the latest G-PCC for color and reflectance attributes while maintaining more efficient encoding and decoding run-times.

Both traditional and learning-based hyperspectral image compression methods suffer from significant quality loss at high compression ratios. To address this, we propose a low-overhead, compression-aware channel filtering method. The encoder derives channel filters via Least Squares Regression between lossy compressed and original images. The bitstream, containing the compressed image and filters, is sent to the decoder, where the filters enhance image quality. This simple, compression-aware approach is compatible with any existing framework, enhancing quality while introducing only a negligible increase in bitstream size and decoding time, thereby achieving low overhead. Experimental results show consistent rate-distortion gains, reducing compression rates by 10.51% to 39.81% on the GF-5 dataset with minimal decoding and storage overhead.

The growing use of hyperspectral images demands efficient compression techniques to handle their extensive spectral data. However, current methods are constrained by their inability to adapt to high bit depth and effectively utilize the spectral characteristics, leading to suboptimal compression ratios. This paper presents a novel hyperspectral compression framework that employs a lightweight spectral super-resolution network to address these limitations. The proposed approach divides the hyperspectral image into two sub-images, comprising two distinct groups of bands: a base image consisting of anchor bands and a supplementary image comprising non-anchor bands. The base image is compressed losslessly using a conventional codec, thereby ensuring the preservation of essential information. In contrast, the supplementary image is compressed efficiently by overfitting a lightweight super-resolution network to predict the non-anchor bands during encoding. The optimized network parameters are encoded as side information to ensure high-quality spectral super-resolution during decoding. Experimental results on the ARAD hyperspectral image dataset demonstrate that our approach significantly outperforms state-of-the-art methods, effectively meeting the demand for efficient hyperspectral image compression while maintaining acceptable processing speeds.

The Graph Fourier Transform (GFT) has achieved remarkable success in point cloud attribute compression due to its adaptability in handling irregular signals. However, the conventional graph-based attribute compression method mostly relies on geometry information to construct the Laplace matrix. In the case of poor geometry-attribute correlation, this method cannot represent the attribute correlation well enough to bring efficient attribute compression performance. In this paper, we propose an efficient point cloud attribute compression scheme using attribute-guided graph Fourier transform. We utilize the attribute features to guide the block partitioning process and design a novel scanning method to efficiently represent attribute patterns, which enhances the efficiency of GFT without imposing extra storage overhead introduced by recording clustering results. Furthermore, a coding module of DC coefficients is proposed for leveraging inter-block correlation, and a tailored coding strategy of AC coefficients is designed based on the distribution of transform coefficients. Experimental results demonstrate that our method achieves better compression performance compared with the latest G-PCC version 22.

A light-weight super-resolution (LSR) method from a single image targeting mobile applications is proposed in this work. LSR predicts the residual image between the interpolated low-resolution (ILR) and high-resolution (HR) images using a self-supervised framework. To lower the computational complexity, LSR does not adopt the end-to-end optimization deep networks. It consists of three modules: 1) generation of a pool of rich and diversified representations in the neighborhood of a target pixel via unsupervised learning, 2) selecting a subset from the representation pool that is most relevant to the underlying super-resolution task automatically via supervised learning, 3) predicting the residual of the target pixel via regression. LSR has low computational complexity and reasonable model size so that it can be implemented on mobile/edge platforms conveniently. Besides, it offers better visual quality than classical exemplar-based methods in terms of PSNR/SSIM measures.

In this paper, a new predictive wavelet transform (PWT) is proposed to solve LiDAR point clouds attribute compression. Our method is a combination of predictive coding and Haar wavelet transform. Based on the spatial information, a hierarchical predictive transform tree is designed to represent 3D irregular data points efficiently. Each level node is classified as a predictive node (P-node) or a transform node (T-node) according to the distances to its adjacent nodes. Then in a top-down coding process, the Haar transform is applied to all T-node pairs, and predictive coding is processed on all P-nodes alternately. It is shown by experimental results that the proposed PWT method offers better R-D performances compared with state-of-the-art methods.

The successive subspace learning (SSL) principle was developed and used to design an interpretable learning model, known as the PixelHop method, for image classification in our prior work. Here, we propose an improved PixelHop method and call it Pixel-Hop++. First, to make the PixelHop model size smaller, we decouple a joint spatial-spectral input tensor to multiple spatial tensors (one for each spectral component) under the spatial-spectral separability assumption and perform the Saab transform in a channel-wise manner, called the channel-wise (c/w) Saab transform. Second, by performing this operation from one hop to another successively, we construct a channel-decomposed feature tree whose leaf nodes contain features of one dimension (1D). Third, these 1D features are ranked according to their cross-entropy values, which allows us to select a subset of discriminant features for image classification. In Pixel-Hop++, one can control the learning model size of fine-granularity, offering a flexible tradeoff between the model size and the classification performance. We demonstrate the flexibility of Pixel-Hop++ on MNIST, Fashion MNIST, and CIFAR-10 three datasets.

Inspired by the recently proposed successive subspace learning (SSL) principles, we develop a successive subspace graph transform (SSGT) to address point cloud attribute compression in this work. The octree geometry structure is utilized to partition the point cloud, where every node of the octree represents a point cloud subspace with a certain spatial size. We design a weighted graph with self-loop to describe the subspace and define a graph Fourier transform based on the normalized graph Laplacian. The transforms are applied to large point clouds from the leaf nodes to the root node of the octree recursively, while the represented subspace is expanded from the smallest one to the whole point cloud successively. It is shown by experimental results that the proposed SSGT method offers better R-D performances than the previous Region Adaptive Haar Transform (RAHT) method.

In this paper, we propose a sphere-projection-based framework for point cloud geometry and attribute lossless and lossy coding. The original point cloud is adaptively divided into blocks, and then we create a fitting sphere in each block for modeling the local geometry structure of the point cloud. Sphere coordination transform and spherical projection scheme are introduced to transfer a 3D point cloud to a set of the range images. A novel compact representation of generated range images based on Morton codes is proposed to separate the range images into occupancy images and attributes vectors for further compression. Experimental results demonstrate that for the LiDAR point clouds datasets in lossless compression, the proposed method offers better performance than geometry-based point cloud compression (G-PCC). For the object point clouds datasets in lossy compression, the proposed method has better rate-distortion (R-D) performance than Draco.

A novel nonparametric machine learning methodology, called successive subspace modeling (SSM), is proposed in this work. Without loss of generality, we use image classification as an illustrative example. The SSM procedure consists of two stages: 1) feature learning and 2) decision learning. For feature learning, we partition input images into overlapping patches of different sizes recursively. While the input images define a vector space, patches of smaller sizes form a sequence of growing subspaces. From the smallest to the largest subspaces, we build a model for each subspace in a successive manner through the Saab transform. At the end, we obtain a lower-dimensional feature vector space that contains significant spatial-spectral information of input images. For decision learning, we summarize the distribution of feature vectors with two techniques; namely, feature space partitioning and local manifold learning. Then, for every test sample, an ensemble decision is made based on decisions at each of its near clusters. The superior performance of SSM is demonstrated on the MNIST dataset.

An ensemble method that fuses the output decision vectors of multiple feedforward-designed convolutional neural networks (FF-CNNs) to solve the image classification problem is proposed in this work. To enhance the performance of the ensemble system, it is critical to increase the diversity of FF-CNN models. To achieve this objective, we introduce diversities by adopting three strategies: 1) different parameter settings in convolutional layers, 2) flexible feature subsets fed into the Fully-connected (FC) layers, and 3) multiple image embeddings of the same input source. Furthermore, we partition input samples into easy and hard ones based on their decision confidence scores. As a result, we can develop a new ensemble system tailored to hard samples to further boost classification accuracy. Experiments are conducted on the MNIST and CIFAR-10 datasets to demonstrate the effectiveness of the ensemble method.

A semi-supervised learning framework using the feedforward-designed convolutional neural networks (FF-CNNs) is proposed for image classification in this work. One unique property of FF-CNNs is that no backpropagation is used in model parameters determination. Since unlabeled data may not always enhance semi-supervised learning [1], we define an effective quality score and use it to select a subset of unlabeled data in the training process. We conduct experiments on the MNIST, SVHN, and CIFAR-10 datasets, and show that the proposed semi-supervised FF-CNN solution outperforms the CNN trained by backpropagation (BP-CNN) when the amount of labeled data is reduced. Furthermore, we develop an ensemble system that combines the output decision vectors of different semi-supervised FF-CNNs to boost classification accuracy. The ensemble systems can achieve further performance gains on all three benchmarking datasets.

An interpretable generative model for handwritten digits synthesis is proposed in this work. Modern image generative models such as the variational autoencoder (VAE) are trained by backpropagation (BP). The training process is complex, and its underlying mechanism is not transparent. Here, we present an explainable generative model using a feedforward design methodology without BP. Being similar to VAEs, it has an encoder and a decoder. For the encoder design, we derive principal-component-analysis-based (PCA-based) transform kernels using the covariance of its inputs. This process converts input images of correlated pixels to uncorrelated spectral components, which play the same role as latent variables in a VAE system. For the decoder design, we convert randomly generated spectral components to synthesized images through the inverse PCA transform. A subject test is conducted to compare the quality of digits generated using the proposed method and the VAE method. They offer comparable perceptual quality yet our model can be obtained at much lower complexity.

One fundamental problem of the convolutional neural network (CNN) is catastrophic forgetting, which occurs when new object classes and data are added while the original dataset is not available any more. Training the network only using the new dataset deteriorates the performance with respect to the old dataset. To overcome this problem, we propose an expanded network architecture, called the ExpandNet, to enhance the CNN incremental learning capability. Our solution keeps filters of the original networks on one hand, yet adds additional filters to the convolutional layers as well as the fully connected layers on the other hand. The proposed new architecture does not need any information of the original dataset, and it is trained using the new dataset only. Extensive evaluations based on the CIFAR -10 and the CIFAR -100 datasets show that the proposed method has a slower forgetting rate as compared to several existing incremental learning networks.

An efficient, scalable and robust approach to the handwritten digits recognition problem based on the Saak transform is proposed in this work. First, multi-stage Saak transforms are used to extract a family of joint spatial-spectral representations of input images. Then, the Saak coefficients are used as features and fed into the SVM classifier for the classification task. In order to control the size of Saak coefficients, we adopt a lossy Saak transform that uses the principal component analysis (PCA) to select a smaller set of transform kernels. The handwritten digits recognition problem is well solved by the convolutional neural network (CNN) such as the LeNet-5. We conduct a comparative study on the performance of the LeNet-5 and the Saak-transform-based solutions in terms of scalability and robustness as well as the efficiency of lossless and lossy Saak transforms under a comparable accuracy level.