Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
Machine understanding : machine perception and machine perception MU
Les Z., Les M., Springer International Publishing, New York, NY, 2020. 215 pp. Type: Book (978-3-030240-69-1)
Date Reviewed: Oct 12 2020

This is a difficult book to define/describe. Some parts are very interesting, presenting good research questions and discussions; other parts lack detail and depth--readers are sometimes just sent to the authors’ previous work--and are hard to read due to the extensive formalism, details, and prerequisites required for understanding.

The authors propose a new view of machine perception and explain their approach and some different ways to implement extended machine perception, that is, machine perception MU (a development of the machine understanding research). Starting with the introduction, it is soon clear that this book is one part of a series of other previously published books and works by the same authors, which is also emphasized by the many chapters where the main references are almost all from the authors’ previous works (chapters 3, 4, 6, 7, and 8). It seems as if readers should be familiar with these previous works before starting this one (and the authors frequently remind readers that some concepts are covered in their other publications and are out of the scope of this book). This may explain the initial negative comments presented in this review, as well as the feeling that something is missing (that is, not included).

The authors insist that classical machine perception is usually regarded as a simple perception and composed of limited pattern recognition processes (for example, object and parts recognition, robot navigation, pedestrian detection), “without paying attention to the cognitive aspects of the perceptual process.” According to them, image perception (visual perception), sound perception, or other sensor-based data (sensorial) processing is usually limited, since human cognition, multisensorial fusion, contextual analysis, and scene understating are much more complex than some low-level pattern recognition tasks. The authors focus on a general scene understanding: lower levels and middle and higher levels of thinking (categories, classes, transformations and generalizations, and hierarchy of concepts, abstractions, and contexts).

Human perception and scene understanding/reasoning are very complex, and the first chapter of this book is dedicated to that: human perception. The interesting chapter 1 presents an overview of different works and theories about human perception. In chapter 2, the authors explain their view of machine perception, from sensory objects, object recognition, and scene analysis, passing by abstractions and categories (for example, line drawing), and going up to concepts formation and machine perception MU research. They then specify that, in this book, they will focus only on visual perception and visual understanding--images, schemes, and drawings, in 2D or 3D.

Chapters 3 through 6 describe the considered classes/categories, object and shape classes, silhouette objects, line drawings, tiling classes, perceptual picture classes, and 2D and 3D. Special attention is paid to perceptual transformations, which include position, orientation, scale, and colors, but also combinations (complex composite class), multi-views, and other sophisticated transformation operators. The authors present a formal description of the classes and transformations (operators). These operators aggregate and allow for higher-level representation of the perceptual objects (in 2D or 3D), allowing for generalization and abstraction in their description. Chapters 7 through 9 describe visual reasoning and these classes/categories and transformations; problem solving and tests that demonstrate the possibilities of the proposed approach; and the visual understanding of a scene and the visual reasoning about the scenes (simple cases).

Machine perception MU, represented in this book by studies in visual understanding and visual reasoning approaches, is one of the most important open problems in this domain of research; however, this book presents only a partial view of this complex “perceptual world,” considering only some simple scenes and also simple object shapes and classes. And I must say that the authors do not really enter into the discussions about machine learning and deep learning for scene analysis, which are among the most cited techniques adopted in this field to do: object detection, object segmentation, object classification, attention mechanisms, scene analysis, movement prediction and object tracking, occlusion treatment, invariance to transformations, 2D and 3D integration, automatic semantic labeling, and so on.

In conclusion, this is an intriguing book, though readers should probably start with the authors’ previous works. Doing so will give readers a better idea of the authors’ understanding and perceptions about this research area.

Reviewer:  Fernando Osorio Review #: CR147080 (2103-0055)
Bookmark and Share
  Reviewer Selected
 
 
Knowledge Representation Formalisms And Methods (I.2.4 )
 
 
Computer Vision (I.5.4 ... )
 
 
Philosophical Foundations (I.2.0 ... )
 
 
General (I.2.0 )
 
Would you recommend this review?
yes
no
Other reviews under "Knowledge Representation Formalisms And Methods": Date
Knowledge representation: an approach to artificial intelligence
Bench-Capon T., Academic Press Prof., Inc., San Diego, CA, 1990. Type: Book (9780120864409)
Jul 1 1991
Truth and modality for knowledge representation
Turner R., MIT Press, Cambridge, MA, 1991. Type: Book (9780262200806)
Nov 1 1991
Constraint relaxation may be perfect
Montanari U., Rossi F. (ed) Artificial Intelligence 48(2): 143-170, 1991. Type: Article
Aug 1 1992
more...

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 1999-2024 ThinkLoud®
Terms of Use
| Privacy Policy