Patent EP0043006A2 successfully refreshed from Google Patents.
EP0043006A2
Expired / Lapsed
EP
⏱️ 0.0 years remaining
Pattern recognition apparatus using prototype probability matching
Assignee: International Business Machines Corp
•
Filing: 1981-06-10
•
Granted: —
•
Expires: 2001-06-10
Abstract
Abstract Pattern recognition apparatus of the kind in which prototype equivalence is determined on the basis of the probability of detected pattern features being present in each prototype wherein each feature probability is determined for all prototypes in a dedicated node (10,-10n) of a parallel feed back network by iteratively selecting the maximum feature probability when individually weighted by previously selected feature probabilities and a relative node displacement weighting. The feature probabilities are determined in relation to elementary segments of an input pattern which is small enough to be deemed invariant under expected distortion and identity relative to a best fit prototype is determined from the final result generated by each node after all elementary segments have been cycled through the network for each prototype.
Inventors
- 👤 James Karl Baker
- 👤 Janet Maciver Baker
Bibliographic Details
Application Number
—
Jurisdiction
EP
Data Source
google_patents
Last Synchronized
2026-10-31 22:47
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/08; G10L15/14; G06; G06F; G06F18/00; G06F18/20; G06F18/29; G06F18/295; G06V; G06V10/00; G06V10/70; G06V10/74; G06V10/75; G06V10/754
Claims Analysis
Total Claims: 7
Claim 1
Independent Claim
74 words
Pattern recognition apparatus of the kind in which prototype equivalence is determined on the basis of the probality of detected pattern features being present in each prototype characterised in that each feature probability is determined for all prototypes in a dedicated node (10 1 -10 n ) of a parallel feed back network by iteratively selecting the maximum feature probability when individually weighed by previously selected feature probabilities and a relative node displacement weighting.
Claim 2
Dependent Claim
74 words
Pattern recognition apparatus of the kind in which prototype equivalence is determined on the basis of the probality of detected pattern features being present in each prototype characterised in that each feature probability is determined for all prototypes in a dedicated node (10 1 -10 n ) of a parallel feed back network by iteratively selecting the maximum feature probability when individually weighed by previously selected feature probabilities and a relative node displacement weighting.
Claim 3
Dependent Claim
179 words
Pattern recognition apparatus arranged to attribute a prototype identity to an input pattern comprising (a) an input stage (12, 14) receiving an input pattern and transmuting it into an ordered sequence of probability factors per preconditioned prototype, outputting each probability factor in parallel to each node (10 1 -10 n ) of (b) a parallel feedback network of a local decision module nodes arranged to select, for each probability factor for each prototype, the maximum value of that factor as weighted by each individual-feedback value of the maximum value selected for the previous factor in the sequence for that prototype as well as by a relative node displacement weighting, and (c) an output stage (FINAL DECISION) responsive to the selected maximum valuesat the nodes for each prototype at the end of the ordered sequence to attribute prototype identity where the probability factors relate to each of a plurality of elementary segments of and spanning the input pattern selected in terms of the worst case prototype so that each elementary segment can be taken to be invariant under expected distortion.
Claim 4
Dependent Claim
179 words
Pattern recognition apparatus arranged to attribute a prototype identity to an input pattern comprising (a) an input stage (12, 14) receiving an input pattern and transmuting it into an ordered sequence of probability factors per preconditioned prototype, outputting each probability factor in parallel to each node (10 1 -10 n ) of (b) a parallel feedback network of a local decision module nodes arranged to select, for each probability factor for each prototype, the maximum value of that factor as weighted by each individual-feedback value of the maximum value selected for the previous factor in the sequence for that prototype as well as by a relative node displacement weighting, and (c) an output stage (FINAL DECISION) responsive to the selected maximum valuesat the nodes for each prototype at the end of the ordered sequence to attribute prototype identity where the probability factors relate to each of a plurality of elementary segments of and spanning the input pattern selected in terms of the worst case prototype so that each elementary segment can be taken to be invariant under expected distortion.
Claim 5
Dependent Claim
125 words
Apparatus as claimed in claim 2 in which each factor and weighting is represented as a logarithmic value, each local decision module including a relative node displacement weighting store (44) and individual feedback gating (42) under control of a common weighting counter (40), the output of the store and the gating being connected to an accumulator memory (52) connected in parallel to a comparator and a register (30), the output of the register being connected to the comparator, the comparator controlling the loading of the register whereby as the counter cycles, the feedback factors and weightings are fed in corresponding pairs to the accumulator memory, added and retained whilst the result is compared with the register conents, the larger value being retained in the register.
Claim 6
Dependent Claim
75 words
Apparatus as claimed in claim 3 in which each local decision module includes a prototype counter (64) driven prototype store (56) outputting the current probability factor for the current segment for each prototype in sequence, the store output being connected to an accumulator (28) connected in turn to a partial result store (32), the output of which is connected to the feedback paths, the prototype counter stepping once per iteration of the common weighting counter.
Claim 7
Dependent Claim
97 words
Apparatus as claimed in claim 4 in which the input stage includes a label store (12e) having a label entry for each node position of each prototype, means (12d) to generate a probability factor in response to each observed elementary segment of an input pattern relative to each label and sequence position and means (12f) to transmit all such factors to the prototype store of each node, and in which each local decision module incorporates a prototype counter driven address mapping arrangement (22, 58, 62) to access the entry appropriate to the node, prototype and sequence position.
References Cited (Backward Citations)
No citation records indexed.
USPTO Annuity & Maintenance Windows
Maintenance schedule is calculated for granted US utility patents.