Image associative memory

Cortadellas, Javier; Amat, Josep

doi:10.1109/icpr.2000.906156

Cited by 4 publications

(5 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Moreover, the memory model does not use recurrent neural networks to recall faces; instead, it stores all faces and their transforms in look-up tables for quick recall. [3,17,18,24,27] This paper leverages existing face detection work.…”

Section: Our Contributionsmentioning

confidence: 99%

“…When presented with a partial or noisy version of one of the images stored on the optical disk, the optical system evolves to a stable state in which those stored images that best match the inputs are temporally locked in the loop. Cortadellas and Amat proposed a facial image associative memory based on silhouettes representation [3]. However, the processing is impossible to distinguish two objects with similar silhouettes but different textures 3 , colors, and other features 4 .…”

Section: Related Workmentioning

confidence: 99%

“…For facial image, image component keys are eyes, nose, and mouth, for example. 3 Image texture is surfaces (the sequence of faces) that can contain data corresponding to not only a color but can be a virtual canvas for a picture or other image. Informally, a texture is an image that can be applied ("texture-mapped") onto an object's surface or particle.…”

Section: Notesmentioning

confidence: 99%

“…Many existing works have attempted to model facial image associative memories with physical/optical devices [17], image pre-processing [3,27], and software modeling [18,24]. The attempts include optical network device implementations, various image filtering and manipulation techniques, recurrent neural network model, and direct imagery pixel-based operations.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Facial Image Associative Memory Model

Tepvorachai

Papachristou

2007

Second NASA/ESA Conference on Adaptive Hardware and Systems (AHS 2007)

View full text Add to dashboard Cite

Facial image associative memory takes a facial input image and returns associated faces pre-embedded in memory. This paper proposes a three-phase implementation process: a) sensory pre-processing, b) information interfusion, and c) association with existing faces. This paper reports on the simulation and performance of the proposed first phase, sensory preprocessing, based on multiple neural network structures to translate image sensory pre-processing into transformed information. The multi-network structure is tested by 46 faces of 21 individuals. The result shows the first phase can produce acceptable associations at 89.1% of all test faces with less than 0.02 energy error at above 0.60 facial image pixel-based correlation (closeness).

show abstract

Section: Our Contributionsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Notesmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Facial Image Associative Memory Model

Tepvorachai

Papachristou

2007

Second NASA/ESA Conference on Adaptive Hardware and Systems (AHS 2007)

View full text Add to dashboard Cite

show abstract

“…Among them, the description based on the region that an arbitrary shape occupies is the simplest one. The benefits of this simple description are that it has a straight meaning for human visual perception systems, it avoids the computation of shape descriptors [3,4,5], which can be ambiguous, and finally, as the description lies on the image domain, it is possible to use specialized hardware architectures like [13] that operate over the image domain at very high speeds.…”

Section: Introductionmentioning

confidence: 99%

Robust Normalization of Shapes

Cortadellas

Amat

Frigola

2002

Discrete Geometry for Computer Imagery

View full text Add to dashboard Cite

The normalization of a binary shape is a necessary step in many image processing tasks based on image domain operations. When one must deal with deformable shapes (due to the projection of non-rigid objects onto the image plane or small changes in the position of the view point), the traditional approaches doesn't perform well. This paper presents a new method for shape normalization based on robust statistics techniques, which allows to keep the location and orientation of shapes constant independent of the possible deformations they can suffer. A numerical comparison of the sensitivity of both methods is used as a measure to validate the proposed technique, together with a ratio of areas between the non-overlapping regions and the overlapping regions of the normalized shapes. The results presented, involving synthetic and real shapes, show that the new normalization approach is much more reliable and robust that the traditional one.

show abstract