Keyang Cheng

Sparse representations based attribute learning for flower classification

Abstract Classification for flowers is a very difficult task. Traditional methods need to built a classifier for each flower category, and obtain large number of flower samples to train these classifiers. In practice, many different types of flowers make the job become very difficult and boring. In this work, we present an attribute based approach for flowers recognition. Particularly, instead of training for a specific category of flowers directly based on manually designed features such as SIFT and HoG, we extract a series of visual attributes from a given set of flower images and generalize these to new images with possibly unknown flowers. A recently proposed sparse representations classification scheme is employed to predict the attributes of a given flower image from any category. In addition, we use a genetic algorithm to find the most discriminative attributes among others for better performance during the stage of flower classification. The effectiveness of the proposed method is validated on a publicly available flower classification database with promising results.

Sparse representations based distributed attribute learning for person re-identification

Searching for specific persons from surveillance videos captured by different cameras, known as person re-identification, is a key yet under-addressed challenge. Difficulties arise from the large variations of human appearance in different poses, and from the different camera views that may be involved, making low-level descriptor representation unreliable. In this paper, we propose a novel Sparse Representations based Distributed Attribute Learning Model (SRDAL) to encode targets into semantic topics. Compared to other models such as ELF, our model performs best by imposing semantic restrictions onto the generation of human specific attributes and employing Deep Convolutional Neural Network to generate features without supervision for attributes learning model. Experimental results show that our method achieves state-of-the-art performance.

A new classifier for facial expression recognition: fuzzy buried Markov model

To overcome the disadvantage of classical recognition model that cannot perform well enough when there are some noises or lost frames in expression image sequences, a novel model called fuzzy buried Markov model (FBMM) is presented in this paper. FBMM relaxes conditional independence assumptions for classical hidden Markov model (HMM) by adding the specific cross-observation dependencies between observation elements. Compared with buried Markov model (BMM), FBMM utilizes cloud distribution to replace probability distribution to describe state transition and observation symbol generation and adopts maximum mutual information (MMI) method to replace maximum likelihood (ML) method to estimate parameters. Theoretical justifications and experimental results verify higher recognition rate and stronger robustness of facial expression recognition for image sequences based on FBMM than those of HMM and BMM.

A New Approach for Facial Expression Recognition Based on Burial Markov Model

To overcome the disadvantage of classical recognition model which cannot perform enough well when there are some noises or lost frames in expression image sequencers, a novel model called burial Markov model is applied in facial expression recognition based on video image sequences. Compared with hidden Markov model, buried Markov model (BMM), as an improved technology of HMM, adds the specific cross-observation dependencies between observation elements in order to increase both accuracy and discriminability. Theoretical justifications and experimental results show that facial expression recognition of video frames based on BMM can get high recognition rate and has strong robustness.

Data-driven pedestrian re-identification based on hierarchical semantic representation: Data-driven Pedestrian Re-identification based on Hierarchical Semantic Representation

Limited number of labeled data of surveillance video causes the training of supervised model for pedestrian re‐identification to be a difficult task. Besides, applications of pedestrian re‐identification in pedestrian retrieving and criminal tracking are limited because of the lack of semantic representation. In this paper, a data‐driven pedestrian re‐identification model based on hierarchical semantic representation is proposed, extracting essential features with unsupervised deep learning model and enhancing the semantic representation of features with hierarchical mid‐level ‘attributes’. Firstly, CNNs, well‐trained with the training process of CAEs, is used to extract features of horizontal blocks segmented from unlabeled pedestrian images. Then, these features are input into corresponding attribute classifiers to judge whether the pedestrian has the attributes. Lastly, with a table of ‘attributes‐classes mapping relations’, final result can be calculated. Under the premise of improving the accuracy of attribute classifier, our qualitative results show its clear advantages over the CHUK02, VIPeR, and i‐LIDS data set. Our proposed method is proved to effectively solve the problem of dependency on labeled data and lack of semantic expression, and it also significantly outperforms the state‐of‐the‐art in terms of accuracy and semanteme.

A novel improved ViBe algorithm to accelerate the ghost suppression

This paper presents a new improved ViBe algorithm approach to accelerate the ghost suppression, which is a robust and efficient background subtraction algorithm for video sequences. The ViBe has the advantages of faster processing speed and lighter computation load compared with other algorithms. For the sake of the real-time performance of the background modeling, it only uses the first frame to build the background model during the process of initialization. However, it will result in introducing ghost area in the subtraction progression, which has an impact on the performance of the background modeling. We put forward a novel method to accelerate the elimination of the ghost area by detecting and reinitializing the region of the ghost area. Our enhanced algorithm is compared with other improved algorithms with the same aim to suppress the generation of the ghost by conducting a series of contrast experiments. The comparison figures show that our method has better performance in ghost suppression. The PCC is used as a metric to evaluate our algorithm performance. The experiment results show that the PCC of our algorithm has improved after the second frame distinctly in contrast to the original algorithm as well as the improved algorithms mentioned in this paper. Besides, the time for processing per frame can still meet the demand of real-time performance.

A new algorithm for pedestrian detection

This article puts forward a novel framework for pedestrian detection tasks, which proposing a model with both sparse reconstruction and class discrimination components, jointly optimized during dictionary learning. We present an efficient pedestrian detection system using mixing sparse features of HOG, FOG and CSS to combine into a Kernel classifier. Results presented on our data set show competitive accuracy and robust performance of our system outperforms current state-of-the-art work.

Pedestrian detection based on kernel discriminative sparse representation

This article puts forward a novel framework for pedestrian detection tasks, which proposing a model with both sparse reconstruction and class discrimination components, jointly optimized during dictionary learning. We present an efficient pedestrian detection system using mixing sparse features of HOG, FOG and CSS to combine into a Kernel classifier. Results presented on our data set show competitive accuracy and robust performance of our system outperforms current state-of-the-art work.

Pedestrian detection based on efficient fused lasso algorithm

The paper presents a new algorithm of pedestrian detection based on efficient fused lasso algorithm (EFLA) and analyses the sparse representation framework. The method considers the structure information of pedestrian image and makes it encoded into the sparse representation model to obtain good discrimination. The proposed algorithm makes use of EFLA to make features that they include the color, texture and shape features extracted from pedestrian image sparse, and detect pedestrian through support vector machine (SVM). The experimental results show that the proposed method has detection nature and better effect on massive data set as well as gives better robustness on the detection of difficult images.