GB1395147A - Feature classifiaction in image analysis - Google Patents

Abstract

1395147 Image analysis IMAGE ANALYZING COMPUTERS Ltd 28 April 1972 [6 May 1971] 13591/71 Heading G1A [Also in in Division G4] In a system for classifying a region of a polished steel (white) surface containing oxide (black) and sulphide (grey) inclusions a video signal is detected against two reference levels as described in Specification 1390938 to provide two signals comprising constant amplitude pulses representing intercepts with oxide and sulphide inclusions respectively, the oxide inclusions being further classified by making two separate measurements on the first signal and comparing these two measurements to identify the kind of oxide inclusion. In Fig. 3 the two detected signals are applied at 10,56 respectively. The sulphide pulses at 56 are accumulated in a store 58 and read out to a counter 62 at the end of each field. The oxide pulses at 10 are accumulated in a store 12 and read out at the end of each field to one of three counters 14, 16, 18 according to the classification. A computer 29 such as is disclosed in Specification 1391056 accumulates values of the total horizontal and vertical projections of the inclusions; at the end of each field the two values are compared by dividing the horizontal by the vertical and comparing the quotient with a constant value K1; a signal X is generated if the quotient is greater than the constant (signifying long thin inclusions of type C) Fig. 2 (not shown). A simple anticoincidence computer 38 such as described in Specification 1264807 accumulates in a counter 40 the total number of inclusions. An agglomerater such as described in U.S. Specification 3795792 accumu lates in a counter 50 the number of closely spaced groups of inclusions. At the end of each field the number of groups is divided by the total number of inclusions and if the quotient is not greater than a second constant K2 a Y signal is issued (signifying a preponderance of goups of inclusions of type B) Fig. 2. According to whether the X or the Y signal is present the oxide pulses are read into counters 14, 16 respectively; if neither X or Y is present (signifying random inclusions of type D-Fig. 2) they are read into counter 18. Individual fields are classified by comparing the counts in accumulators 12, 58 and separate counts are made of the number of fields classified as predominately of each type A, B, C or D. The counters 14, 16, 18, 62 may accumulate pulses over say 500 small fields of a sample to arrive at a classification for the whole sample sheet. A number of alternative schemes for sub-classifying the oxide inclusions are mentioned in the Specification.