Salient Shape Research Articles

Vision-Action Semantic Associative Learning Based on Spiking Neural Networks for Cognitive Robot

Establishing cognitive environments help cognitive robots understand human actions, languages and observed objects. In this paper, a cognitive robotic model based on a novel spiking bidirectional associative memory (BAM) method is presented to establish a cognitive environment. The spiking BAM network uses a supervised spike-based learning rule to learn the relationship between the semantic information of vision and action, where a new coding scheme is proposed to encode and decode vision and action semantic information. Vision semantic information is obtained by the deeply salient shape detection (DSSD) method. Action semantic information is obtained by a spiking neural network (SNN) with the proposed coding method. Experimental results demonstrate that the spiking BAM achieves a good convergence ability and relatively high recall accuracy. Specially, for the cognitive development of the robot, this system can be used to improve the robot’s ability to infer human intentions in its natural interactions with humans.

Search mode, not the attentional window, determines the magnitude of attentional capture

A salient color distractor is known to capture attention during search for a less salient shape target, but the mechanisms underlying attentional capture are debated. Theeuwes (2004, Psychonomic Bulletin & Review, 11(1), 65–70) argued that attentional capture depends on the size of the attentional window. If the attentional window is large, search is efficient and attentional capture should be stronger because the distractor is more likely to be inside the window. Consistently, we found higher search efficiency and more attentional capture in singleton than in feature search. However, differences in attentional capture only occurred when singleton and feature search were performed by different groups of participants, but not when singleton and feature search occurred unpredictably in the same group of participants. This result contradicts the attentional window account because search efficiency was always higher in singleton than in feature search. Rather, the results support search mode theory, which claims that participants looked for the most salient stimulus in singleton search (“singleton detection mode”), which resulted in more capture by the salient color distractor. When search types varied unpredictably, it was impossible to apply a consistent search strategy, which eliminated the differences between singleton and feature search. Further, we manipulated search efficiency by target–nontarget similarity. With dissimilar nontargets, the target was salient and search efficiency was high. Therefore, the attentional window account predicts more capture. However, we found the opposite result in singleton search and no difference in feature search. Taken together, these observations are inconsistent with the attentional window account but support search mode theory.

Violated expectations for spatial and feature attributes of visual trajectories modulate event-related potential amplitudes across the visual processing hierarchy

During visual perception, the brain must combine its predictions about what is to be perceived with incoming relevant information. The present study investigated how this process interacts with attention by using event-related potentials that index these cognitive mechanisms. Specifically, this study focused on examining how the amplitudes of the N170, N2pc, and N300 would be modulated by violations of expectations for spatial and featural attributes of visual stimuli. Participants viewed a series of shape stimuli in which a salient shape moved across a set of circular locations so that the trajectory of the shape implied the final position and shape of the stimulus. The final salient stimuli occurred in one of four possible outcomes: predictable position and shape, predictable position but unpredictable shape, unpredictable position but predictable shape, and unpredictable position and shape. The N170 was enhanced by unpredictable positions and shapes, whereas the N300 was enlarged only by unpredictable positions. The N2pc was not modulated by violations of expectations for shapes or positions. Additionally, it was observed post-hoc that the P1pc amplitude was increased by unpredictable shapes. These findings revealed that incorrect prediction increases neural activity. Furthermore, they suggest that prediction and attention interact differently in different stages of visual perception, depending on the type of attention being engaged: The N170 indexes initial prediction error signalling irrespective of the type of information (spatial or featural) in which error occurs, followed by the N300 as a marker of prediction updating involving reorientation of spatial attention.

Identifying and extracting bark key features of 42 tree species using convolutional neural networks and class activation mapping

The significance of automatic plant identification has already been recognized by academia and industry. There were several attempts to utilize leaves and flowers for identification; however, bark also could be beneficial, especially for trees, due to its consistency throughout the seasons and its easy accessibility, even in high crown conditions. Previous studies regarding bark identification have mostly contributed quantitatively to increasing classification accuracy. However, ever since computer vision algorithms surpassed the identification ability of humans, an open question arises as to how machines successfully interpret and unravel the complicated patterns of barks. Here, we trained two convolutional neural networks (CNNs) with distinct architectures using a large-scale bark image dataset and applied class activation mapping (CAM) aggregation to investigate diagnostic keys for identifying each species. CNNs could identify the barks of 42 species with > 90% accuracy, and the overall accuracies showed a small difference between the two models. Diagnostic keys matched with salient shapes, which were also easily recognized by human eyes, and were typified as blisters, horizontal and vertical stripes, lenticels of various shapes, and vertical crevices and clefts. The two models exhibited disparate quality in the diagnostic features: the old and less complex model showed more general and well-matching patterns, while the better-performing model with much deeper layers indicated local patterns less relevant to barks. CNNs were also capable of predicting untrained species by 41.98% and 48.67% within the correct genus and family, respectively. Our methodologies and findings are potentially applicable to identify and visualize crucial traits of other plant organs.

PeaGlyph: Glyph design for investigation of balanced data structures

For many data analysis tasks, such as the formation of well-balanced groups for a fair race or collaboration in learning settings, the balancing between data attributes is at least as important as the actual values of items. At the same time, comparison of values is implicitly desired for these tasks. Even with statistical methods available to measure the level of balance, human judgment, and domain expertise plays an important role in judging the level of balance, and whether the level of unbalance is acceptable or not. Accordingly, there is a need for techniques that improve decision-making in the context of group formation that can be used as a visual complement to statistical analysis. This paper introduces a novel glyph-based visualization, PeaGlyph, which aims to support the understanding of balanced and unbalanced data structures, for instance by using a frequency format through countable marks and salient shape characteristics. The glyph was designed particularly for tasks of relevance for investigation of properties of balanced and unbalanced groups, such as looking-up and comparing values. Glyph-based visualization methods provide flexible and useful abstractions for exploring and analyzing multivariate data sets. The PeaGlyph design was based on an initial study that compared four glyph visualization methods in a joint study, including two base glyphs and their variations. The performance of the novel PeaGlyph was then compared to the best “performers” of the first study through evaluation. The initial results from the study are encouraging, and the proposed design may be a good alternative to the traditional glyphs for depicting multivariate data and allowing viewers to form an intuitive impression as to how balanced or unbalanced a set of objects are. Furthermore, a set of design considerations is discussed in context of the design of the glyphs.

Narrative Elicitation as Ethnography: Methodological Insights From the Examination of Children's Perspective Marking in Amdo Tibetan.

This paper employs a case study with Amdo Tibetan children to demonstrate the benefits of narrative elicitation for ethnographic language socialization research in under-studied languages. Primarily by examining spontaneous verbal interaction, existing language socialization research has demonstrated how salient grammatical resources shape children's understanding of cultural belief systems pertaining to sociality and the appropriate display of emotion. However, spontaneous data do not always capture children's full linguistic repertoires and competencies, and may therefore present a partial picture of their mastery over particular grammatical systems. One such area that remains to be studied is how children use interactional cues to build their emerging knowledge of grammatical perspective marking in Tibetan languages. This paper integrates narrative elicitation with ethnographic methods from language socialization to examine how Amdo Tibetan children mark perspective using evidentiality, the grammatically-obligatory encoding of knowledge source, an area not frequently documented in language socialization studies. Language socialization research involved 15-months of participant observation, audio-video recording, and analysis of spontaneous interactions with children aged 1–4. This ethnographic research found that adults' narratives highlighted local theories about the importance of compassion (Tib. snying rje) by using grammatical evidentiality to emphasize characters' direct experiences in the story-world. However, grammatical evidentiality was under-represented in children's spontaneous talk. To provide further insight into children's mastery of evidentiality in this culturally salient communicative genre, I conducted narrative elicitation tasks with seven Amdo Tibetan children, aged 2–7. By framing narrative elicitation tasks as forums for social interaction in family homes, I adapted a method traditionally used in experimentation to complement the study of naturalistic interaction. Interaction analysis of the elicited narratives found that family members positioned young children as novice narrators, leading to dialogic rather than monologic narratives. Young children co-constructed shared perspectives on narrated events, and used evidentiality in conventionalized ways by mirroring the grammatical forms of adults' previous utterances. By adapting narrative elicitation tasks to language socialization's ethnographic methods, this paper models how qualitative researchers can locate patterns in children's experiences of language across complementary settings of data collection, an endeavor that is particularly important to research with child speakers of under-documented languages.

Design and Optimization of Coaxial Reluctance Magnetic Gear With Different Rotor Topologies

This study investigates the influence of inner reluctance rotor topologies configurations on the electromagnetic performances of coaxial reluctance magnetic gears (CRMGs). It considers flux barrier (FB) rotor topology, as well as the salient pole (SP) rotor. Both SP and FB topologies are optimized by means of different salient shapes and novel supershape formula, respectively. The CRMG performances in terms of torque density and cogging torque have been investigated and optimized fairly by using 3-D finite element analysis. With an appropriate rotor topology, FB-CRMG can achieve higher torque capacity and lower cogging torque in comparison with SP-CRMG topologies. The prototype of optimized FB-CRMG has been built to validate the predictions of output torque quality, its loss, and efficiency.

End‐to‐end feature fusion Siamese network for adaptive visual tracking

According to observations, different visual objects have different salient features in different scenarios. Even for the same object, its salient shape and appearance features may change greatly from time to time in a long-term tracking task. Motivated by them, we proposed an end-to-end feature fusion framework based on Siamese network, named FF-Siam, which can effectively fuse different features for adaptive visual tracking. The framework consists of four layers. A feature extraction layer is designed to extract the different features of the target region and search region. The extracted features are then put into a weight generation layer to obtain the channel weights, which indicate the importance of different feature channels. Both features and the channel weights are utilized in a template generation layer to generate a discriminative template. Finally, the corresponding response maps created by the convolution of the search region features and the template are applied with a fusion layer to obtain the final response map for locating the target. Experimental results demonstrate that the proposed framework achieves state-of-the-art performance on the popular Temple-Color, OTB50 and UAV123 benchmarks.

Frequency of exposure and target salience affect the extinction of value-driven attention capture

Our work has demonstrated that previously rewarded features that remain task relevant continue to capture attention during an extinction phase. Here, we investigated the conditions under which such value-driven capture is extinguished. In three experiments (E), participants learned to associate a target color with reward, and another with no reward in a visual search paradigm. One week later, during the extinction phase, this association was no longer reinforced. In E1, the colors that previously defined targets were presented as distractors, as is typical of paradigms in this literature, however we failed to observe an effect of former reward-associations on RT, suggesting that reward did not reliably capture attention when task-irrelevant. E2 was identical to E1, but included an absent condition where neither formerly rewarded or non-rewarded colors were present. In E2 a value-driven capture effect was not observed. However, trials containing either a previously rewarded or non-rewarded distractor that had previously been selected were significantly slower compared to trials where no previously selected stimulus was present, which persisted throughout extinction. These findings demonstrate that the role of selection history (previous experience selecting a feature) appears to play a critical role in shaping attention, unique from that of reward history. E3 was also identical to E1 but targets were defined as salient shape singletons during extinction, another common feature in this literature. In contrast to E1 and E2, value-driven attention capture by formerly reward associated colors was extinguished. The rate of extinction in E3 was rapid, which could be the result of competition for attentional priority between the salient singleton and the distractor. These findings demonstrate that the extinction of value-driven capture depends on various factors that affect priority for attention, especially those features that bias attention away from reward-associated features.

The “Sh-Ape Bias” in Non-Linguistic Categorization: Comparisons between Children and Other Apes

ABSTRACTChildren show a bias toward information about shape when labeling or determining category membership for novel objects. The body of work with human children suggests that the shape bias is not restricted to linguistic contexts but is highly contingent on task demands. Testing nonhumans could provide additional information about the salience of shape cues in the absence of linguistic relevance. In order to assess the salience of shape versus color using an identical task in children and apes, we presented two adult zoo-housed chimpanzees (Pan troglodytes) and 56 three–five-year-old children with a relational matching task in which samples and comparison stimuli matched on either shape or color. Whereas children of all ages performed above chance on the task, chimpanzees performed at close to chance levels overall. However, closer inspection revealed that, whereas children performed better on shape (86%) versus color trials (78.5%), chimpanzees showed the opposite pattern, performing at chance on shape trials (49%) and above chance on color trials (72%). Children also made quicker responses on shape versus color trials, whereas chimpanzees showed the opposite pattern. Whereas shape is a highly salient cue for Western children, color may be a more salient natural cue in nonhuman primates’ natural environments. Thus, the shape bias does not appear to be an evolutionarily ancient bias when assigning category membership.

Statistical learning shapes distractor suppression

Even though traditionally attentional control is considered to be the result of the interaction between top-down and bottom-up mechanisms, Awh, Belopolsky and Theeuwes (2012, TICS) have suggested a theoretical framework in which this division is argued to be incomplete. They argued that the history of attentional deployments can elicit lingering selection biases, which are unrelated to top-down goals or the physical salience of items. Whereas previous work has primarily focused on target selection, here we investigated whether implicitly learned statistical regularities can influence distractor processing. We used the additional singleton task in which participants search for a salient shape singleton while ignoring a color distractor singleton. This color distractor was systematically presented more often in one location than in all other locations. Using measures of behavior, we demonstrate that for this high-probability location, both the amount of attentional capture by distractors and the efficiency of selecting the target were reduced. Measures of brain activity, specifically the Pd (distractor positivity) and N2pc components of the event-related potential, were used to track the allocation of attention and suppression to lateralized positions in the arrays. We find evidence for a Pd-like component for the high probability location, even though both the target and the distractor were presented on the vertical meridian (top and bottom) and thus could not themselves elicit a lateralized component in the ERP. When the distractor was presented at the lateralized high probability location, this also generated a Pd, while a distractor presented at a low probability location generated an N2pc. The same was found for when the target was presented at the high versus low lateralized probability location. We interpret these findings as evidence that spatial statistical regularities influence distractor processing through inhibition of high probability locations.

Salience drives overt selection of two equally relevant visual targets.

In the present study, we investigated whether salience determines the sequence of selection when participants search for two equally relevant visual targets. To do this, attentional selection was tracked overtly as observers inspected two items of differing physical salience: one a highly salient color singleton, and the other a less salient shape singleton. Participants were instructed to make natural eye movements in order to determine whether two line segments contained within the two singletons were oriented in the same or in different directions. Because both singleton items were task-relevant, participants had no reason to inspect one item before the other. As expected, observers fixated both targets on the majority of trials. Critically, saccades to the color singleton preceded saccades to the less salient shape singleton on the majority of trials. This demonstrates that the order of attentional object selection is largely determined by stimulus salience when task relevance is equated.

The influence of psychological factors on postpartum weight retention 12 months post-birth

BackgroundDuring the first postpartum year 20% of women retain excessive weight from pregnancy (postpartum weight retention; PPWR), which predicts long-term overweight/obesity.ObjectiveThe aim of this study was to explore the associations between psychological factors (depression, anxiety and stress symptoms and body attitudes) in late gestation and at 12-months postpartum with PPWR one-year post-birth.MethodsPregnant women (N = 176) completed questionnaires in early–mid pregnancy (Time 1; mean (SD) = 16.97 (1.35) weeks), late pregnancy (Time 2; mean (SD) = 33.33 (2.05) weeks), and one year postpartum (Time 3; mean (SD) = 53.12 (3.34) weeks). Women provided demographic characteristics, height and pre-pregnancy weight at Time 1. At Times 2 and 3, weight, depressive, anxiety and stress symptoms and body attitudes (salience of weight and shape, attractiveness, feeling fat, and strength and fitness) were assessed in addition to physiological, socio-contextual and lifestyle factors. Gestational weight gain and PPWR were calculated. Hierarchical linear regression models were conducted to explore variance in 12-month PPWR.ResultsOverall, models explained 26–39% variance in PPWR. Gestational weight gain in late pregnancy and low attractiveness at 12 months postpartum were the only variables associated significantly with 12-month PPWR.ConclusionWhile psychological factors did not appear to be important direct contributors to PPWR at 12 months, the overall contribution of all variables suggests that such factors may be implicated in a small and incremental way. Exploration of the interactions between variables will help unpack potential mechanisms of the development of PPWR at 12 months post-birth.

Hierarchical Vertex Regression-Based Segmentation of Head and Neck CT Images for Radiotherapy Planning.

Segmenting organs at risk from head and neck CT images is a prerequisite for the treatment of head and neck cancer using intensity modulated radiotherapy. However, accurate and automatic segmentation of organs at risk is a challenging task due to the low contrast of soft tissue and image artifact in CT images. Shape priors have been proved effective in addressing this challenging task. However, conventional methods incorporating shape priors often suffer from sensitivity to shape initialization and also shape variations across individuals. In this paper, we propose a novel approach to incorporate shape priors into a hierarchical learning-based model. The contributions of our proposed approach are as follows: 1) a novel mechanism for critical vertices identification is proposed to identify vertices with distinctive appearances and strong consistency across different subjects; 2) a new strategy of hierarchical vertex regression is also used to gradually locate more vertices with the guidance of previously located vertices; and 3) an innovative framework of joint shape and appearance learning is further developed to capture salient shape and appearance features simultaneously. Using these innovative strategies, our proposed approach can essentially overcome drawbacks of the conventional shape-based segmentation methods. Experimental results show that our approach can achieve much better results than state-of-the-art methods.

A new method for retrieving batik shape patterns

Batik as a traditional art is well regarded due to its high aesthetic quality and cultural heritage values. It is not uncommon to reuse versatile decorative shape patterns across batiks. General‐purpose image retrieval methods often fail to pay sufficient attention to such a frequent reuse of shape patterns in the graphical compositions of batiks, leading to suboptimal retrieval results, in particular for identifying batiks that use copyrighted shape patterns without proper authorization for law‐enforcement purposes. To address the lack of an optimized image retrieval method suited for batiks, this study proposes a new method for retrieving salient shape patterns in batiks using a rich combination of global and local features. The global features deployed were extracted according to the Zernike moments (ZMs); the local features adopted were extracted through curvelet transformations that characterize shape contours embedded in batiks. The method subsequently incorporated both types of features via matching a weighted bipartite graph to measure the visual similarity between any pair of batik shape patterns through supervised distance metric learning. The derived similarity metric can then be used to detect and retrieve similar shape patterns appearing across batiks, which in turn can be employed as a reliable similarity metric for retrieving batiks. To explore the usefulness of the proposed method, the performance of the new retrieval method is compared against that of three peer methods as well as two variants of the proposed method. The experimental results consistently and convincingly demonstrate that the new method indeed outperforms the state‐of‐the‐art methods in retrieving salient shape patterns in batiks.

Statistical regularities modulate attentional capture.

The present study investigated whether statistical regularities can influence visual selection. We used the classic additional singleton task in which participants search for a salient shape singleton while ignoring a color distractor singleton. The color distractor singleton was systematically presented more often in 1 location than in all other locations. For this high-probability location, we found that both the amount of attentional capture by distractors and the efficiency of selecting the target were reduced. There was a spatial gradient of suppression, as the attentional capture effect and the efficiency of selecting the target scaled with the distance from the high-probability location. Some participants were aware of the statistical regularities, but this did not affect the results whatsoever. We interpret these findings as evidence that spatially statistical regularities that are unknown to the observer can influence attention such that locations that have a high probability of containing a distractor are suppressed relative to all other locations. (PsycINFO Database Record

The relationship between antenatal body attitudes, pre-pregnancy body mass index, and gestational weight gain

Objectivesan estimated 50% of women experience excessive gestational weight gain (GWG). Maternal body attitudes are associated with GWG, however this relationship is complex and may differ based on pre-pregnancy body mass index (BMI) or gestational age. The aim of this study was to explore the moderating role of maternal pre-pregnancy BMI on the relationship between body attitudes in early-to-mid and late pregnancy and GWG. Design/Participantspregnant women less than 18 weeks gestation were recruited for a postal questionnaire study via Australian pregnancy online forums, pregnancy and parenting magazines, and antenatal clinics. In early-mid pregnancy (Time 1; mean (M) = 16.81 weeks gestation, standard deviation (SD) = 1.18), participants reported demographics, pre-pregnancy weight, height, and body attitudes (salience of weight and shape, attractiveness, strength and fitness and feeling fat). In late pregnancy, body attitudes (Time 2; M = 32.65 weeks gestation, SD = 0.91) and weight (Time 3; M = 37.15 weeks gestation, SD = 1.55) were reported. Pre-pregnancy BMI and total GWG were calculated. Moderation analyses were conducted. Findingsin early-mid pregnancy, pre-pregnancy BMI moderated the relationship between feeling fat and GWG. Pre-pregnancy BMI did not moderate the relationship between body attitudes and GWG for salience of weight and shape, attractiveness or strength and fitness in early-mid pregnancy. In late pregnancy, pre-pregnancy BMI moderated the relationship between all four body attitude facets (salience of weight and shape, attractiveness, feeling fat and strength and fitness) and GWG. Conclusion/Implications for practicethe relationship between body attitudes and GWG was moderated by pre-pregnancy BMI, particularly in late pregnancy. It is recommended that antenatal health care providers monitor women's body attitudes throughout pregnancy to aid in the management of healthy GWG and promote positive maternal and infant health outcomes. This is particularly important for women entering pregnancy with an underweight/normal weight BMI.

Analyzing the Decorative Style of 3D Heritage Collections Based on Shape Saliency

As technologies for 3D acquisition become widely available, it is expected that 3D content documenting heritage artifacts will become increasingly popular. Nevertheless, to provide access to and enable the creative use of this content, it is necessary to address the challenges to its access. These include the automatic enrichment of 3D content with suitable metadata so that content does not get lost. To address these challenges, this article presents research on developing technologies to support the organization and discoverability of 3D content in the Cultural Heritage (CH) domain. This research takes advantage of the fact that heritage artifacts have been designed throughout the centuries with distinctive design styles. Hence, the shape and the decoration of an artifact can provide significant information on the history of the artifact. The main contributions of this article include an ontology for documenting 3D representations of heritage artifacts decorated with ornaments such as architectural mouldings. In addition, the article presents a complementary shape retrieval method based on shape saliency to improve the automatic classification of the artifact’s semantic information based on its 3D shape. This method is tested on a collection of Regency ornament mouldings found in domestic interiors. This content provides a rich dataset on which to base the exploration of issues common to many CH artifacts, such as design styles and decorative ornament.

Dust particle detection in surveillance video using salient visual descriptors

Outdoor surveillance video that suffers from dirty camera lenses has the potential for deteriorated performance in many applications, such as vehicle tracking and target recognition. This paper proposes to identify the dust particles in images using a set of salient visual descriptors. More specifically, the proposed approach exploits an extended feature descriptor comprising localized color, texture and shape saliency features. These proposed features are further incorporated into a machine learning approach, followed by a dust particle localization approach, for detecting dust particle in images. The proposed approach is able to achieve superior dust particle detection performance to that of conventional approaches, as evaluated in real-world video surveillance scenarios.

Saliency detection via joint modeling global shape and local consistency

Saliency detection is the task of locating informative regions in an image, which is a challenging task in computer vision. In contrast to the existing saliency detection models that focus on either local or global image property, an effective salient object detection method is introduced based on joint modeling global shape and local consistency. To this end, Restricted Boltzmann Machine (RBM) is utilized to model salient object shape as global image property and Conditional Random Field (CRF), on the other hand, is adopted to achieve its local consistency. In order to obtain the final saliency map, a universal framework is introduced to combine the results of RBM and CRF. Experimental results on five benchmark datasets demonstrate that the proposed saliency detection method performs favorably against the existing state-of-the-art algorithms.