Abstract Image sensors are increasingly being used in biodiversity monitoring, with each study generating many thousands or millions of pictures. Efficiently identifying the species captured by image is a critical challenge for advancement this field. Here, we present an automated identification method wildlife pictures remote camera traps. Our process starts images that cropped out background. We then use improved sparse coding spatial pyramid matching (ScSPM), which extracts dense SIFT...

A proper strategy to alleviate overfitting is critical a deep neural network (DNN). In this paper, we introduce the cross-loss-function regularization for boosting generalization capability of DNN, which results in multi-loss regularized DNN (ML-DNN) framework. For particular learning task, e.g., image classification, only single-loss function used all previous DNNs, and intuition behind multiloss framework that extra loss functions with different theoretical motivations (e.g., pairwise...

Detection of mitotic tumor cells per tissue area is one the critical markers breast cancer prognosis. The aim this paper to develop a method for automatic detection figures from histological slides using partially supervised deep learning framework. Unlike previous literature, which has focused on solving problem mitosis in weakly annotated datasets centroid pixel labels (weak labels) only without taking advantage available pixel-level (strong other datasets, paper, we design novel framework...

The Gaussian process (GP) latent variable model (GPLVM) has the capability of learning low-dimensional manifold from highly nonlinear data high dimensionality. As an unsupervised dimensionality reduction (DR) algorithm, GPLVM been successfully applied in many areas. However, its current setting, is unable to use label information, which available for tasks; therefore, researchers proposed kinds extensions order utilize extra among supervised (SGPLVM) shown better performance compared with...

Inspired by the recent image feature learning work, we propose a novel key point detection approach for object tracking. Our can select mid-level interest points max pooling over local descriptor responses from set of filters. Linear filters are first learned targets in frames. Then is performed data driven spatial supporting field to detect discriminant points, and thus detected bear higher level semantic meanings, which apply tracking structured matching. We show that our system robust...

This paper introduces a feature descriptor called Shape of Gaussian (SOG), which is based on general design framework Signal Probability Density Function (SOSPDF). SOSPDF takes the shape signal's probability density function (pdf) as its feature. Under such view, both histogram and region covariance often used in computer vision are features. Histogram describes by discrete approximation way. Region an incomplete parameterized multivariate distribution. Our proposed SOG full Gaussian, so it...

In this paper, we study discriminative analysis of symmetric positive definite (SPD) matrices on Lie groups (LGs), namely, transforming an LG into a dimension-reduced one by optimizing data separability. particular, take the space SPD matrices, e.g., covariance as concrete example LGs, which has proved to be powerful tool for high-order image feature representation. The transformation is achieved within-class compactness well between-class separability based popular graph embedding...

Modern efficient Convolutional Neural Networks(CNNs) always use Depthwise Separable Convolutions(DSCs) and Architecture Search(NAS) to reduce the number of parameters computational complexity. But some inherent characteristics networks are overlooked. Inspired by visualizing feature maps N×N(N>1) convolution kernels, several guidelines introduced in this paper further improve parameter efficiency inference speed. Based on these guidelines, our parameter-efficient CNN architecture, called...

This paper presents an ordered-patch-based image classification framework integrating the Grassmannian manifold to address handwritten digit recognition, face and scene recognition problems. Typical methods explore appearances without considering spatial causality among distinctive domains in image. To issue, we introduce representation use autoregressive moving average (ARMA) model characterize representation. First, each is encoded as a sequence of ordered patches, both local appearance...

Feature extraction is a crucial part of computer vision. In this paper, we present novel method that can automatically extract relevant features from video for action recognition and identity human who makes the action, in single framework. We propose watermark embedding to represent as 2-D wavelet transform. The feature consists Deep Belief Network (DBN) on Discrete Fourier Transforms (DFTs) tracked video. then use activations trained network inputs non-linear Support Vector Machine (SVM)...

The authors introduce the concept of a spline resampling in particle filter to deal with high accuracy and sample impoverishment. is usually based on linear transformation weights particles, so it affects filtering accuracy. consists two parts: spread states. former particles obtain highly accurate filtering, latter point states prevent impoverishment due decline diversity hypothesis after resampling. Two transformations are sequentially implemented incorporate each other. Then, propose...

Dimensionality reduction (DR) has been considered as one of the most significant tools for data analysis. One type DR algorithms is based on latent variable models (LVM). LVM-based can handle preimage problem easily. In this paper we propose a new model, named thin plate spline model (TPSLVM). Compared to well-known Gaussian process (GPLVM), our proposed TPSLVM more powerful especially when dimensionality space low. Also, robust shift and rotation. This investigates two extensions TPSLVM,...

In this paper, a fast and robust video copy detection scheme is proposed, which suitable for the DCT-coded sequences. To address efficiency effectiveness issue, we extract signature directly from compressed domain. The sequence clusters are constructed with fixed length. Each cluster consists of several fictional key-frames. For each key-frame, some low-middle frequency full DCT coefficients obtained block coefficients, their ordinal measure computed acts as signature. A rotation...