A novel machine learning and compressive sensing (CS) based super-resolution (SR) algorithm for the restoration of remote images is proposed in this paper. This new relies on idea that high-resolution (HR) image patches can be correctly recovered from downsampled low-resolution (LR) under two mild conditions, i.e., sparsity patches, incoherence between projection matrix. Consequently if most HR represented as a sparse linear combination elements dictionary incoherent with matrix, accurately...

White blood cell (WBC) detection is one of the most basic and key steps in automatic WBC recognition system. Its accuracy stability greatly affect whole This paper presents a novel method for based on boundary support vectors (BSVs). Firstly, v-Support Vector Regression (v-SVR) introduced. Then sparse BSVs are obtained while fitting 1D histogram by v -SVR. Next so-needed threshold value directly sifted from these limited vectors. Finally entire connective regions segmented original image....

Crowd counting is an important research topic in the fields of computer vision and image processing, with monitoring management crowded scenes becoming increasingly prominent issue.Existing methods still suffer from problem severe overlap density maps within dense areas, leading to inadequate localization accuracy.This paper presents innovative on crowd localization.Firstly, addressing limitations performance existing algorithms, we optimize generation method FIDT maps, decoupling tasks.By...

A new algorithm is proposed for TV commercial detection from color video in this paper. High shot change frequency and "still" shots of the trademark information, which are two basic characteristics commercial, exploited to distinguish commercials general programs. Our retrieval system based on text realized through a slide window. First, histogram difference computed consecutive images then four common transitions including cut, dissolve, fade in/out wipe detected. maximum gradient...

Combination of gray water and land SAR image wavelet texture information, present a new segmentation method surface. Firstly, extracting level co-occurrence matrix the sub-blocks image, then using transform to extract norm average deviation as feature information sub-image; Accordingly, two types establish suitable combination separation measure multi-dimensional space; Finally, K-means clustering algorithm segment image. The experimental results show that effect is better than common method.

Background extraction is an important step in vehicle detection. In the actual scene, change of illumination will lead to a tremendous background change. It necessary update model reasonably and effectively as changes. order solve this problem, paper proposes adaptive ViBe model. Firstly, two kinds detection errors their corresponding error function are defined. Then, according range these errors, set reasonable evaluation conditions determined adjust unreasonable threshold value, which...

In asymmetric retrieval systems, models with different capacities are deployed on platforms computational and storage resources. Despite the great progress, existing approaches still suffer from a dilemma between efficiency accuracy due to limited capacity of lightweight query model. this work, we propose an Asymmetric Feature Fusion (AFF) paradigm, which advances systems by considering complementarity among features just at gallery side. Specifically, it first embeds each image into various...

Object detection in remote-sensing images is an important and challenging task. With the development of deep learning technology, method based on convolutional neural network has made considerable progress. However, due to problems images, such as dense arrangement, arbitrary direction complex background, traditional networks are difficult use adequately semantic information images. We design a novel single-stage detector feature fusion three-branch attention. The map extracted by backbone...

Linear feature detection is very important in computer vision, image segmentation and pattern recognition. The drawbacks of the standard Hough transform (SHT) are quantization error location line-segments. In this paper, nonuniform HT parameter proposed to decrease influence uniform line-segments detection. Moreover, transforming relative decomposing digital line different scale segments an increase veracity. Experimental results included show that method can achieve high accuracy has...

Feature extraction is a key step in the classification and recognition problem. Features from different methods vary lot with separability their feature space. We propose novel method based on distance matrix to evaluate by describing in-class aggregation between-class scatter of every class. Finally each class measured individually. Experiments synthetic data ORL face dataset prove its effectiveness advantage regard conventional methods.

In this paper, a Bayesian network-based assessment model used for evaluating the innovation of cloud computing industry is presented. Firstly, measurement industrial clusters designed. Then network and self-learning method to are proposed. Finally, accompanying with empirical data, most likely status value key variables influencing can be predicted. This provide theory basis researching innovative development industries.

Semantic understanding of images remains an important research challenge for the image and video retrieval community. A novel natural scene method based on non-negative sparse coding is proposed in this paper. It firstly combines with spatial pyramid matching feature extraction representation. Then, coding, it ranks Euclidean distances from query to each K-nearest neighbors database. With help SIFT flow label transfer, we finally realize segmentation recognition images. The experimental...