Background Dataset Research Articles

Extreme data compression for the CMB

We apply the Karhunen-Lo\'eve methods to cosmic microwave background (CMB) data sets, and show that we can recover the input cosmology and obtain the marginalized likelihoods in $\mathrm{\ensuremath{\Lambda}}$ cold dark matter cosmologies in under a minute, much faster than Markov chain Monte Carlo methods. This is achieved by forming a linear combination of the power spectra at each multipole $l$, and solving a system of simultaneous equations such that the Fisher matrix is locally unchanged. Instead of carrying out a full likelihood evaluation over the whole parameter space, we need evaluate the likelihood only for the parameter of interest, with the data compression effectively marginalizing over all other parameters. The weighting vectors contain insight about the physical effects of the parameters on the CMB anisotropy power spectrum ${C}_{l}$. The shape and amplitude of these vectors give an intuitive feel for the physics of the CMB, the sensitivity of the observed spectrum to cosmological parameters, and the relative sensitivity of different experiments to cosmological parameters. We test this method on exact theory ${C}_{l}$ as well as on a Wilkinson Microwave Anisotropy Probe (WMAP)-like CMB data set generated from a random realization of a fiducial cosmology, comparing the compression results to those from a full likelihood analysis using CosmoMC. After showing that the method works, we apply it to the temperature power spectrum from the WMAP seven-year data release, and discuss the successes and limitations of our method as applied to a real data set.

Mixed model and stability analysis of spring wheat genotype yield evaluation data from Manitoba, Canada

In western Canada and in many agricultural areas around the world, new crop genotypes are evaluated over a number of locations and years in multi-environment trials (MET) to investigate yield, yield stability, agronomic, and quality characteristics, with the ultimate goal to predict future genotype performance in commercial fields. This evaluation informs decisions about the commercial value of new crop genotypes, with a primary user of this information being farmers. Currently in many regions of Canada as the first step in analysis of this MET data, values usually are expressed as a percentage of a designated check genotype value at each site-year (trial), usually followed by a relatively simplistic statistical analysis of this percentage data. There are a number of problems with this traditional approach including selection of an appropriate check genotype or genotypes, and the necessary consistent performance of the check genotype over a number of locations and years. Following the recent approach of other countries and jurisdictions, MET spring wheat genotype yield data (kg ha−1) that had been collected from 2000 to 2009 from various locations in Manitoba, Canada were subjected to mixed model statistical analysis. The results of the mixed model analysis compared very favourably to the historical traditional approach, and proved to be superior in situations such as a specific year in the dataset (2007) when the designated check genotype performed anomalously poorly. These results indicated that as little as five trial sites in a single year provided sufficient data for reliable prediction of a new genotype’s yield performance, given a background dataset comprised of approximately 45 spring wheat genotypes tested over eight years. The wheat genotype yield data also was subjected to estimation of several different stability measures to investigate differences in yield stability between genotypes in the dataset. Results indicated relatively stable yield performance for most genotypes over a range of site-years (environments).

Life cycle assessment of nickel products

To support the data requirements of stakeholders, the Nickel Institute (NI) conducted a global life cycle impact assessment (LCIA) to show, with indicators, the potential environmental impacts of the production of nickel and ferronickel from mine to refinery gate. A metal industry wide agreed approach on by-products and allocation was applied. Nine companies, comprising 19 operations, contributed data, representing 52 % of global nickel metal production and 40 % of global ferronickel production. All relevant pyro- and hydrometallurgical production routes were considered, across most major nickel-producing regions. Data from Russia, the biggest nickel-producing nation, was included; the Chinese industry did not participate. 2011 was chosen as reference year for data collection. The LCIA applied allocation of impacts of by-products using both economic and mass allocations. A sensitivity analysis was conducted to further understand the relevance and impact of the different allocation approaches. The primary extraction and refining steps are the main contributors to primary energy demand (PED) and global warming potential (GWP), contributing 60 and 70 % to the PED for the production of 1 kg class I nickel and 1 kg nickel in ferronickel, respectively, and over 55 % of the GWP for both nickel products. The PED for 1 kg class 1 nickel was calculated to be 147 MJ, whilst the PED for 1 kg nickel in ferronickel was calculated to be three times higher at 485 MJ. The main factors influencing energy demand in the metallurgical processes are ore grade and ore mineralogy. Sulphidic ore is less energy intensive to process than oxidic ore. Eighty-six percent of the production volume from class 1 nickel producers, in this study, is from sulphidic ore. All ferronickel was produced from oxidic ore. The LCIA results, including a sensitivity analysis of the impact of producers with higher and lower PED, reflect the influence of the production route on energy demand and on environmental impact categories. Conformant to relevant ISO standards, and backed-up with a technical and critical review, this LCIA quantifies the environmental impacts associated with the production of the main nickel products. With this study, a sound background dataset for downstream users of nickel has been provided. The Nickel Institute aims to update their data in the coming years to reflect upon changes in technology, energy efficiency, and raw material input.

Using maximum entropy to predict the potential distribution of an invasive freshwater snail

SummaryEcological niche modelling is a technique used to estimate potential distributions of invasive species based on available occurrence data and associated environmental conditions. Maximum entropy (Maxent) is a powerful method for ecological niche modelling and yet has only rarely been applied to aquatic species.Here we applied Maxent to estimate the potential distribution of the invasive Chinese mystery snail (Cipangopaludina chinesis) in Wisconsin and analysed several methodological issues associated with using Maxent for lentic species.To generate Maxent estimates of the potential distribution ofC. chinesis, we used presence records from 292 lakes, five spatially explicit climatic variables, and two lake‐specific environmental data sets (area, conductivity) from 7995 lakes. Our investigations included three aspects that could affect model prediction accuracy and transferability: (i) combining climate and lake variables into a single data set in two different ways, using each lake as a single observation and as a grid of 1 ha cells; (ii) varying the size of the background data set (locations without presences); and (iii) contrasting environmental conditions between locations with and withoutC. chinesis.The lake‐based model had higher accuracy than the grid‐based model, although both models had accuracy values indicative of good performance. Conductivity and lake area were important predictor variables for both models, but had higher contribution to the lake‐based model accuracy. Decreasing the background sample size minimally affected model accuracy and thus Maxent can be used even when background sampling does not meet the algorithm's default settings. Lastly, lakes that were environmentally dissimilar from lakes with knownC. chinesisrecords were more likely to be predicted unsuitable by both grid‐based and lake‐based models.Overall, the models predicted high potential suitability across Wisconsin lakes forC. chinesis, especially in lakes ≥60 ha. Our study provides evidence that small or environmentally biased presence data sets may underestimate the number of environmentally suitable locations of invasive species.

Regional scale hydrogeochemical mapping of the northern Yilgarn Craton, Western Australia: a new technology for exploration in arid Australia

The northern Yilgarn Craton, with an extensive mineral exploration history and relatively fresh and neutral groundwaters, was selected to test the utility of regional hydrogeochemical mapping in Australia. The assembled data of 2509 groundwater samples (generally at 4–8 km spacing) are relatively unbiased, allowing robust statistical analysis such as testing sample types (flowing v. ‘stagnant’), contamination, and lithological controls on groundwater characteristics. Lithological indicators were developed to map underlying bedrock through cover. Areas with discrepancies between groundwater results and previous geological mapping were identified. Where these are areas previously discounted as prospective for mineral commodities, they may now be re-considered on this basis. Even in well explored parts of this region, this study identified new areas which may have prospective rocks overlain with a thin (<50 m) veneer of granitic material. A large background data set was produced that has significant benefits for lithological discrimination, mineral exploration, guiding human and livestock drinking water supplies and environmental management (e.g. mine closure). Groundwater chemistry can effectively map large-scale lithological changes in these semi-arid environments, and in turn can reduce uncertainly about the prospectivity of areas within the northern Yilgarn Craton. This should reduce the drilling and associated costs required to delineate a target. The methods and interpretation developed in this study will enhance mineral exploration into covered environments as much of the northern two thirds of Australia has similar groundwater environments. This methodology can be expanded into covered arid terrains worldwide. Additionally, this can be used as background to improve interpretation of other small scale studies. Improving the exploration potential of other more difficult regions of Australia will encourage industry exploration in Australia, and provide potential future economic gains.

Carotid dual-energy CT angiography: Evaluation of low keV calculated monoenergetic datasets by means of a frequency-split approach for noise reduction at low keV levels

Background and purposeCalculated monoenergetic ultra-low keV datasets did not lead to improved contrast-to-noise ratio (CNR) due to the dramatic increase in image noise. The aim of the present study was to evaluate the objective image quality of ultra-low keV monoenergetic images (MEIs) calculated from carotid DECT angiography data with a new monoenergetic imaging algorithm using a frequency-split technique. Materials and methods20 patients (12 male; mean age 53±17 years) were retrospectively analyzed. MEIs from 40 to 120keV were reconstructed using the monoenergetic split frequency approach (MFSA). Additionally MEIs were reconstructed for 40 and 50keV using a conventional monoenergetic (CM) software application. Signal intensity, noise, signal-to-noise ratio (SNR) and CNR were assessed in the basilar, common, internal carotid arteries. ResultsUltra-low keV MEIs at 40keV and 50keV demonstrated highest vessel attenuation, significantly greater than those of the polyenergetic images (PEI) (all p-values <0.05). The highest SNR level and CNR level was found at 40keV and 50keV (all p-values <0.05). MEIs with MFSA showed significantly lower noise levels than those processed with CM (all p-values <0.05) and no significant differences in vessel attenuation (p>0.05). Thus MEIs with MFSA showed significantly higher SNR and CNR compared to MEIs with CM. ConclusionCombining the lower spatial frequency stack for contrast at low keV levels with the high spatial frequency stack for noise at high keV levels (frequency-split technique) leads to improved image quality of ultra-low keV monoenergetic DECT datasets when compared to previous monoenergetic reconstruction techniques without the frequency-split technique.

Optimization of Background Information and Layer Thickness for Improved Accuracy of Water-Vapor Profile Retrieval from Ground-Based Microwave Radiometer Measurements at K-Band

Ground-based microwave radiometers operating at frequencies near the 22.235 GHz (K-band) water vapor absorption line have been used extensively for remote sensing of water vapor in the troposphere, both the integrated amount and its profile. This paper explores the potential to use ground-based, zenith-pointing K-band radiometer measurements along with optimized background data sets consisting of radiosonde profiles to detect dynamic changes and gradients in water vapor profiles. To explore this capability, the HUMidity EXperiment 2011 (HUMEX11) was conducted at the U.S. Department of Energy’s (DOE) Atmospheric Radiation Measurement (ARM) Southern Great Plains (SGP) Site near Lamont, OK, USA. This enables the choice of appropriate retrieval parameters to monitor temporal changes in atmospheric water vapor profiles. The results of this study illustrate that in a retrieval algorithm both the choice of the size of the background data set measured near the radiometer measurement time and the choice of atmospheric layer thickness affect the ability to remotely sense dynamic changes in water vapor. In general, it is found that background data sets of larger size provide better accuracy in a statistical sense but inhibit the ability to detect gradients.

Semantic Image Segmentation with Contextual Hierarchical Models

Semantic segmentation is the problem of assigning an object label to each pixel. It unifies the image segmentation and object recognition problems. The importance of using contextual information in semantic segmentation frameworks has been widely realized in the field. We propose a contextual framework, called contextual hierarchical model (CHM), which learns contextual information in a hierarchical framework for semantic segmentation. At each level of the hierarchy, a classifier is trained based on downsampled input images and outputs of previous levels. Our model then incorporates the resulting multi-resolution contextual information into a classifier to segment the input image at original resolution. This training strategy allows for optimization of a joint posterior probability at multiple resolutions through the hierarchy. Contextual hierarchical model is purely based on the input image patches and does not make use of any fragments or shape examples. Hence, it is applicable to a variety of problems such as object segmentation and edge detection. We demonstrate that CHM performs at par with state-of-the-art on Stanford background and Weizmann horse datasets. It also outperforms state-of-the-art edge detection methods on NYU depth dataset and achieves state-of-the-art on Berkeley segmentation dataset (BSDS 500).

Video parsing via spatiotemporally analysis with images

Effective parsing of video through the spatial and temporal domains is vital to many computer vision problems because it is helpful to automatically label objects in video instead of manual fashion, which is tedious. Some literatures propose to parse the semantic information on individual 2D images or individual video frames, however, these approaches only take use of the spatial information, ignore the temporal continuity information and fail to consider the relevance of frames. On the other hand, some approaches which only consider the spatial information attempt to propagate labels in the temporal domain for parsing the semantic information of the whole video, yet the non-injective and non-surjective natures can cause the black hole effect. In this paper, inspirited by some annotated image datasets (e.g., Stanford Background Dataset, LabelMe, and SIFT-FLOW), we propose to transfer or propagate such labels from images to videos. The proposed approach consists of three main stages: I) the posterior category probability density function (PDF) is learned by an algorithm which combines frame relevance and label propagation from images. II) the prior contextual constraint PDF on the map of pixel categories through whole video is learned by the Markov Random Fields (MRF). III) finally, based on both learned PDFs, the final parsing results are yielded up to the maximum a posterior (MAP) process which is computed via a very efficient graph-cut based integer optimization algorithm. The experiments show that the black hole effect can be effectively handled by the proposed approach.

USING A TRIANGULAR MATCHING APPROACH FOR LATENT FINGERPRINT AND PALMPRINT IDENTIFICATION

The identification of fingerprints and palmprints is considered a challenging research line in Biometrics. Nowadays, the accuracy of these techniques highly depends on the quality of the involved impressions, specially if the matching is performed in latent cases. In the present work, a new algorithm that reports a high accuracy in both cases is presented. This proposal requires a minimum amount of manually marked information in latent impressions and deals successfully with problems of missing and spurious minutiae. Moreover, we improved a verification algorithm based on a previously introduced feature model. The algorithm uses a strategy for finding adaptable local matches between substructures obtained from images. The experimental results show that our proposal achieves high accuracy for all cases, despite the major differences that exist between a palmprint and a fingerprint. For latent fingerprint identification our approach shows its robustness in retrieving 258 latents from different size scale of the background dataset, achieving a rank-1 identification rate over 57% in all cases. Carrying out a similar experimentation for latent palmprint identification, our approach achieved a rank-1 identification rate over 75%.

A new fast approach to nonparametric scene parsing

Scene parsing is a challenging research area in computer vision. It provides a semantic label for each pixel in image. Most scene parsing approaches are parametric based which need a model that is acquired through a learning stage. In this paper, a new nonparametric approach to scene parsing is proposed which does not require a learning stage. All introduced nonparametric approaches are based on patch correspondence. Our proposed method does not require explicit patch matching which makes it fast and effective. The proposed approach has two parts. In the first part, a new generative approach to transfer semantic labels from a training image to an unlabelled test image is proposed. To do this, a graphical model is constructed over regions of both the training and test images. Then, based on the proposed graphical model, a quadratic convex function is defined on likelihood probability of each region. Cost function is defined such that contextual information and object-level information are both considered. In the second part of our approach, by using the proposed method of transfer knowledge, a new nonparametric scene parsing approach is given. To evaluate the proposed approach, it is applied on the MSRC-21, Stanford background, LMO, and SUN datasets. The obtained results show that our approach outperforms comparable state-of-the-art nonparametric approaches.

Effective background data selection for SVM-based speaker recognition with unseen test environments: more is not always better

This study focuses on formulating a procedure to select effective negative examples for the development of improved Support Vector Machine (SVM)-based speaker recognition. Selection of a background dataset, or a collection of negative examples, is the crucial step for building an effective decision surface between a target speaker and the non-target speakers. Previous studies heuristically fixed the number of negative examples used based on available development data for performance evaluation; nevertheless, in real applications this does not guarantee sustained performance for unseen data, as will be shown. In the proposed model selection framework, a novel ranking method is first exploited to rank order the negative examples for selecting a set of background datasets with various population sizes. Next, an error estimation and model-selection criterion are proposed and employed to select the most suitable target model among the model candidates. The experimental validation, conducted on the NIST SRE-2008 and SRE-2010 data, demonstrates that the proposed background data selection slightly but consistently outperforms the fixed-size background data selection, and achieves a relative improvement of +6 % over the non-selection background framework in terms of minDCF.

Amino Acid Changes in Disease-Associated Variants Differ Radically from Variants Observed in the 1000 Genomes Project Dataset

The 1000 Genomes Project data provides a natural background dataset for amino acid germline mutations in humans. Since the direction of mutation is known, the amino acid exchange matrix generated from the observed nucleotide variants is asymmetric and the mutabilities of the different amino acids are very different. These differences predominantly reflect preferences for nucleotide mutations in the DNA (especially the high mutation rate of the CpG dinucleotide, which makes arginine mutability very much higher than other amino acids) rather than selection imposed by protein structure constraints, although there is evidence for the latter as well. The variants occur predominantly on the surface of proteins (82%), with a slight preference for sites which are more exposed and less well conserved than random. Mutations to functional residues occur about half as often as expected by chance. The disease-associated amino acid variant distributions in OMIM are radically different from those expected on the basis of the 1000 Genomes dataset. The disease-associated variants preferentially occur in more conserved sites, compared to 1000 Genomes mutations. Many of the amino acid exchange profiles appear to exhibit an anti-correlation, with common exchanges in one dataset being rare in the other. Disease-associated variants exhibit more extreme differences in amino acid size and hydrophobicity. More modelling of the mutational processes at the nucleotide level is needed, but these observations should contribute to an improved prediction of the effects of specific variants in humans.

Labeling Spain With Stanford

We present an end-to-end framework for outdoor scene region decomposition, learned on a small set of randomly selected images that generalizes well to multiple data sets containing images from around the world. We discuss the different aspects of the framework especially a generalized variational inference method with better approximations to the true marginals of a graphical model. Experimentally, we explain why the framework is robust and performs competitively on many diverse scene data sets, including several unseen scene types. We have obtained high pixel-level accuracies ( ≈ 80%) in three of the four data sets, which include a benchmark data set known as the Stanford background data set. Our model obtained over 70% accuracy on the fourth data set, which contained a number of indoor and close-up images that are significantly different from our training examples.

Efficient semantic image segmentation with multi-class ranking prior

Semantic image segmentation is of fundamental importance in a wide variety of computer vision tasks, such as scene understanding, robot navigation and image retrieval, which aims to simultaneously decompose an image into semantically consistent regions. Most of existing works addressed it as structured prediction problem by combining contextual information with low-level cues based on conditional random fields (CRFs), which are often learned by heuristic search based on maximum likelihood estimation. In this paper, we use maximum margin based structural support vector machine (S-SVM) model to combine multiple levels of cues to attenuate the ambiguity of appearance similarity and propose a novel multi-class ranking based global constraint to confine the object classes to be considered when labeling regions within an image. Compared with existing global cues, our method is more balanced between expressive power for heterogeneous regions and the efficiency of searching exponential space of possible label combinations. We then introduce inter-class co-occurrence statistics as pairwise constraints and combine them with the prediction from local and global cues based on S-SVMs framework. This enables the joint inference of labeling within an image for better consistency. We evaluate our algorithm on two challenging datasets which are widely used for semantic segmentation evaluation: MSRC-21 dataset and Stanford Background dataset and experimental results show that we obtain high competitive performance compared with state-of-the-art methods, despite that our model is much simpler and efficient.

Which functionalities are available in the electronic health record systems used by French general practitioners? An assessment study of 15 systems

ObjectiveWhereas an unprecedented effort is currently under way worldwide for the implementation of electronic health record (EHR) systems, their capabilities are poorly understood, especially in primary care. The objective of this study was to assess the main functionalities of the EHR systems used in French general practices. MethodsAmong the 20 EHR systems marketed in France, we assessed the 15 systems used by more than 1500 general practitioners in the Provence-Alpes-Côte d’Azur region in the southeast part of France. Each EHR system was assessed in a general practice office, using two clinical vignettes describing virtual patient consultations. The evaluation criteria were derived from the EuroRec requirements for EHR system quality. The assessment scale included 37 criteria grouped into three sets: background data, consultation data, and exchange functionalities. The scoring system used, totalling 64 points, was based on the validation of the criteria and was adjusted based on the possibility of standardising the data. A high score indicated a good EHR system quality. ResultsThe median global score was 32 points out of a possible 64 (range: 20–39). The median score was 12 points out of 22 (range: 6–15) for the background data set, 16 points out of 32 (range: 9–22) for the consultation data set, and four points out of 10 (range: 0–6) for the exchange functionalities. No association was found between the number of users and the assessment score of the EHR systems (p=0.79). One third of the EHR systems lacked a problem list and only one of them supported the episode of care. ConclusionFunctionalities noticeably vary among the EHR systems currently used in French primary care. Whereas these systems are globally very focused on drug prescriptions, several core functionalities are frequently lacking. They are also poorly interoperable for healthcare professionals and patients. Further research is necessary to assess their actual use.

SpADS: An R Script for Mass Spectrometry Data Preprocessing before Data Mining

The recent application of Mass Spectrometry (MS) to Nucleic Acid Programmable Protein Array (NAPPA) technique for proteins identification by non-classical methods leads to the needs of more sophisticated algorithm for peak recognition. NAPPA technique allows for functional proteins to be synthesized in situ directly from printed cDNAs but faces the difficulty generated by the presence of master mix and lysate molecules peaks appearing as background in the overall spectra. A wide range of tools are available to analyze proteins conventional mass spectra corresponding to few molecular species. None of them is optimized for background subtraction. Moreover, peak identification is performed by statistical analysis on characteristics peaks and thus background subtraction can alter outcome by erasing characteristic peaks. A first attempt to overcome the so far discussed problem is here discussed. The result of this effort is the development of SpADS: Spectrum Analyzer and Data Set manager-an R script for MS data preprocessing-therein discussed. SpADS provides useful preprocessing functions such binning and peak extractions, as available tools, and provides functions of spectra background subtraction and dataset managing. It is entirely developed in R, thus free of charge. A cluster k means implementation is here used to improve results of SpADS preprocessing on test datasets and on NAPPA expressed proteins.

The importance of correcting for sampling bias in MaxEnt species distribution models

AbstractAimAdvancement in ecological methods predicting species distributions is a crucial precondition for deriving sound management actions. Maximum entropy (MaxEnt) models are a popular tool to predict species distributions, as they are considered able to cope well with sparse, irregularly sampled data and minor location errors. Although a fundamental assumption of MaxEnt is that the entire area of interest has been systematically sampled, in practice, MaxEnt models are usually built from occurrence records that are spatially biased towards better‐surveyed areas. Two common, yet not compared, strategies to cope with uneven sampling effort are spatial filtering of occurrence data and background manipulation using environmental data with the same spatial bias as occurrence data. We tested these strategies using simulated data and a recently collated dataset on Malay civet Viverra tangalunga in Borneo.LocationBorneo, Southeast Asia.MethodsWe collated 504 occurrence records of Malay civets from Borneo of which 291 records were from 2001 to 2011 and used them in the MaxEnt analysis (baseline scenario) together with 25 environmental input variables. We simulated datasets for two virtual species (similar to a range‐restricted highland and a lowland species) using the same number of records for model building. As occurrence records were biased towards north‐eastern Borneo, we investigated the efficacy of spatial filtering versus background manipulation to reduce overprediction or underprediction in specific areas.ResultsSpatial filtering minimized omission errors (false negatives) and commission errors (false positives). We recommend that when sample size is insufficient to allow spatial filtering, manipulation of the background dataset is preferable to not correcting for sampling bias, although predictions were comparatively weak and commission errors increased.Main ConclusionsWe conclude that a substantial improvement in the quality of model predictions can be achieved if uneven sampling effort is taken into account, thereby improving the efficacy of species conservation planning.

Objective algorithm to separate signal from noise in a Poisson-distributed pixel data set

A statistical method to determine the background level and separate signal from background in a Poisson-distributed background data set is described. The algorithm eliminates the pixel with the highest intensity value in an iterative manner until the sample variance equals the sample mean within the estimated uncertainties. The eliminated pixels then contain signal superimposed on the background, so the integrated signal can be obtained by summation or by a simple extension by profile fitting depending on the user's preferences. Two additional steps remove `outliers' and correct for the underestimated extension of the peak area, respectively. The algorithm can be easily modified to specific needs, and an application on crystal truncation rods is presented, dealing with a sloping background.

Integrating low-level and semantic features for object consistent segmentation

The aim of semantic segmentation is to assign each pixel a semantic label. Numerous methods for semantic segmentation have been proposed in recent years and most of them chose pixel or superpixel as the processing primitives. However, as the information contained in a pixel or a superpixel is not discriminative enough, the outputs of these algorithms are usually not object consistent. To tackle this problem, we introduce the concept of object-like regions as a new and higher level processing primitive. We first experimentally showed that using groundtruth segments as processing primitives can boost semantic segmentation accuracy, and then proposed a novel method to produce regions that resemble the groundtruth regions, which we named them as ‘object-like regions’. We achieve this by integrating state of the art low-level segmentation algorithms with typical semantic segmentation algorithms through a novel semantic feature feedback mechanism. We present experimental results on the publicly available image understanding dataset MSRC21 and stanford background dataset, showing that the new method can achieve relatively good semantic segmentation results with far fewer processing primitives.