{"title":"Hierarchical and modular attention","authors":"H. Wechsler","doi":"10.1109/CAMP.1995.521044","DOIUrl":"https://doi.org/10.1109/CAMP.1995.521044","url":null,"abstract":"The flow of visual input reaching the eye consists of huge amounts of time-varying information. It is crucial for both biological vision and automated systems to perceive and comprehend such a constantly changing environment within a relatively short processing time. To cope with such a computational challenge, one should locate and analyze only the information relevant to the current task by quickly focusing on selected areas of the scene as needed. Attention makes perception computationally tractable and helps with tasks such as object recognition. Attention permeates the whole stream of visual computation, it is both hierarchical and modular, and it involves representations, processing and strategies. Attentional mechanisms are intimately related to adaptation processes, and high-level attention corresponds to competitive, functional and learned behavioral programs. Attention consists of both data- and model-driven processes and their relationships, and it covers several levels such as sensory, reactive and behavioral processes. An example of how attention can be implemented considers time-varying imagery and it shows how functional linked pyramids and zoom lens operations lead to the generation of visual saccades. Both the time-varying imagery and the corresponding recognition memory are organized as pyramids and uniform indexing and classification interfaces using an attention pyramid are established. This paper concludes with a discussion on promising venues for future research that are most likely to enhance our understanding of attentional mechanisms.","PeriodicalId":277209,"journal":{"name":"Proceedings of Conference on Computer Architectures for Machine Perception","volume":"177 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1995-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115219944","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"An object oriented computational model for handwritten document recognition","authors":"K. Ohmori","doi":"10.1109/CAMP.1995.521036","DOIUrl":"https://doi.org/10.1109/CAMP.1995.521036","url":null,"abstract":"This paper describes a new computational model for a handwritten document recognition system. It consists of a perceptive subsystem that recognizes each character image extracted from a document using a template matching method and a cognitive subsystem that recognizes a series of input character images as a sentence using semantical and syntactical knowledge. Semantical and syntactical knowledge is represented in a concept graph. Receiving character recognition results, the cognitive subsystem specializes general knowledge so that the concept graph represents a recognition result for the document. An object oriented model is used to specialize general knowledge by means of creating an instance from a class. In cases where some characters are recognized incorrectly, abduction is carried out by means of inference from other character recognition results. The document recognition system will be realized by a parallel object oriented model and suitable for massive parallel processing.","PeriodicalId":277209,"journal":{"name":"Proceedings of Conference on Computer Architectures for Machine Perception","volume":"63 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1995-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115636493","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Video-rate pyramid optical flow computation on the linear SIMD array IVIP","authors":"M. Johannesson, Mats Gokstorp","doi":"10.1109/CAMP.1995.521051","DOIUrl":"https://doi.org/10.1109/CAMP.1995.521051","url":null,"abstract":"The IVIP (IR Video Image Processor) signal processor array is used to process a video sequence to obtain its optical flow. Optical flow is the apparent motion of the intensity patterns in image sequences, and it can be used to obtain information about scene structure and object motion. A differential-based method for optical flow using second-order derivatives is implemented. We investigate both one-level and pyramid implementations. A pyramid implementation makes it possible to estimate much larger image velocities than a one-level implementation. IVIP is a row-parallel bit-serial SIMD processor array. The special ALU design which incorporates a serial-parallel multiplier/divider together with good interprocessor communication makes it very effective for large convolutions. In this paper, we show that we can process 512/sup 2/ image sequences at 50 Hz frame rate for one-level flow and 25 Hz for multi-level pyramid flow.","PeriodicalId":277209,"journal":{"name":"Proceedings of Conference on Computer Architectures for Machine Perception","volume":"134 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1995-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131282749","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Efficient image processing applications on a network of workstations","authors":"M. Hamdi, Chi-kin Lee","doi":"10.1109/CAMP.1995.521033","DOIUrl":"https://doi.org/10.1109/CAMP.1995.521033","url":null,"abstract":"Using a cluster of networked workstations as an inexpensive parallel computational platform is an appealing idea. However, very little is known about modeling their parallel performance since most of the the developed models have been designed with traditional parallel computers in mind. In this paper we model the performance of this computing environment for synchronous parallel iterative algorithms. One specific algorithm of this class that is treated in detail in this paper is the parallel image processing convolution. Our model takes into consideration the communication capability of the network, the computing capabilities of the workstations, and load imbalance among the workstations. It was shown that our models accurately model the performance of synchronous iterative algorithms on a cluster of workstations. Moreover, this model can be used to tune various parameters in the system to enhance its performance.","PeriodicalId":277209,"journal":{"name":"Proceedings of Conference on Computer Architectures for Machine Perception","volume":"16 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1995-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121334824","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Satellite Digital Elevation Model on the heterogeneous OPENVISION parallel computer","authors":"H. Essafi, C. Mazzoni, P. Julien, O. Jamet","doi":"10.1109/CAMP.1995.521072","DOIUrl":"https://doi.org/10.1109/CAMP.1995.521072","url":null,"abstract":"The goal of the Digital Elevation Model is to generate an accurate three-dimensional scene using a stereo vision technique. In the stereo matching process two techniques are utilized, an area-based and feature-based to generate a disparity map. In our application we use an area-based approach coupled with the prediction validation techniques. The computation of the Digital Elevation Model (DEM) is based on correlation, also called matching, to determine the pixel correspondence in a pair of stereo spatial images. It is a fundamental step in digital mapping. The French \"Institut Geographique National\" (IGN) has developed a system to provide DEM. The kernel of this system is based on an incremental correlation method, which is the bottleneck in the map production because of its expenditure of computing time. In the same way the CEA-LETI has developed in collaboration with the IRIT laboratory (Toulouse University), a SIMD calculator SYMPATI2 dedicated to image processing, and integrated in the OPENVISION real-time system. The IGN DEM of SPOT images (6000/spl times/6000) takes 20 hours using a Sparc 10 workstation. In order to reduce this computation time we studied the parallelization and the implementation the IGN algorithm on OPENVISION.","PeriodicalId":277209,"journal":{"name":"Proceedings of Conference on Computer Architectures for Machine Perception","volume":"26 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1995-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130850176","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Visual attention: detecting abrupt onsets within the selective tuning model","authors":"John K. Tsotsos, Sean M. Culhane, W. Y. Wai","doi":"10.1109/CAMP.1995.521022","DOIUrl":"https://doi.org/10.1109/CAMP.1995.521022","url":null,"abstract":"The paper focuses on one dimension of a model of visual attention, namely the detection and quantification of abrupt onsets and offsets. The overall model is based on the concept of selective tuning. The goal of the research is to develop a model of visual attention that has both biological plausibility as well as computational utility. Abrupt onsets are well known attention capture cues and play a large role not only in signaling salient events in everyday life, but also figure prominently in most psychophysical experimental paradigms. The solution is simple, easily parallelized, yields excellent performance, and provides useful robot head control cues for onset foveation. The model is described in some detail and several performance examples are shown. A description of the implementation is also included.","PeriodicalId":277209,"journal":{"name":"Proceedings of Conference on Computer Architectures for Machine Perception","volume":"56 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1995-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133347114","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Systolic processors applied to computer vision systems","authors":"F. Serratosa, P. Millán, E. Montseny","doi":"10.1109/CAMP.1995.521037","DOIUrl":"https://doi.org/10.1109/CAMP.1995.521037","url":null,"abstract":"Low level image processing needs to apply simple operations to large sets of data. This processing has to be done quickly. Systolic processors enable this to be done. This study describes the design of a reconfigurable systolic processor for the iconic processing of images in real time (video rate). The computers of the processor (basic cells, delay elements, and interconnection network) are presented paying special attention to the basic cell. The basic operations used in the iconic processing of images are studied. The cell is designed so that it can carry out these operations. In the results section, the configurations of the processor for some preprocessing algorithms (convolution, median filtering, narrow edge extraction) are presented and the typical performance of the processor and the techniques used (segmentation and parallelism) are also shown.","PeriodicalId":277209,"journal":{"name":"Proceedings of Conference on Computer Architectures for Machine Perception","volume":"70 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1995-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131966719","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Design, implementation, and performance of a scalable multi-camera interactive video capture system","authors":"Martin Frankel, Jon, Webb","doi":"10.1109/CAMP.1995.521029","DOIUrl":"https://doi.org/10.1109/CAMP.1995.521029","url":null,"abstract":"We describe a system for interactive, real time, multi-camera video capture and display. The system is almost entirely software-based and as a result is flexible and expandable. Computation and data storage is provided by the Carnegie Mellon-Intel Corporation iWarp computer. Video display is accomplished using a High Performance Parallel Interface network to write to a high resolution frame buffer Video can be displayed as it is captured, with only a single frame latency. We provide interactivity with a VCR-like graphical interface running on a host workstation, which in turn controls the operation of the capture system. This system allows a user to interactively monitor, capture, and replay video with the ease of use of a VCR, yet with flexibility and performance that is unavailable in all but the most expensive and customized video capture systems.","PeriodicalId":277209,"journal":{"name":"Proceedings of Conference on Computer Architectures for Machine Perception","volume":"118 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1995-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122464302","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Fast computation of multiscalar symmetry in foveated images","authors":"M. Bolduc, G. Sela, M. Levine","doi":"10.1109/CAMP.1995.521013","DOIUrl":"https://doi.org/10.1109/CAMP.1995.521013","url":null,"abstract":"This paper discusses two components of a Robot Eye intended as an active vision system to be mounted on a mobile robot. The first component is a foveated vision sensor which is based on an overlapping receptive field model for data reduction. We present the adapted scan-line algorithm used to compute so-called retinal images and a description of the implementation of the system on a network of DSP's. The second component computes salient points in the foveated image and is motivated by the biological processes which guide primate gaze fixation. The model of attention and its real-time implementation are described. Experimental results obtained with these algorithms are also presented.","PeriodicalId":277209,"journal":{"name":"Proceedings of Conference on Computer Architectures for Machine Perception","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1995-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130931761","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Towards the empirical design of massively parallel arrays for spatially mapped applications","authors":"M. Herbordt, C. Weems","doi":"10.1109/CAMP.1995.521020","DOIUrl":"https://doi.org/10.1109/CAMP.1995.521020","url":null,"abstract":"Although SIMD arrays have been built since the 1960's, they have undergone few empirical studies. The underlying problems-which have included the lack of a unified architectural framework and the computational intractability of simulating large PE arrays-are addressed through the use of trace compilation, a novel approach to trace driven simulation. The results indicate the benefits of adding another level to current SIMD array memory designs. Also, surprising results were obtained about performance effects of varying cache associativity and block size. Together, they indicate that while SIMD array programs have sufficient locality to make PE caches worthwhile, the type of locality may differ fundamentally from that of serial machine and multiprocessor programs. Other results demonstrate the limitations of increasing the datapath width and inter PE communication bandwidth without corresponding improvements in other processor features.","PeriodicalId":277209,"journal":{"name":"Proceedings of Conference on Computer Architectures for Machine Perception","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1995-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128496907","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}