Abstract: Text is recognized in a scanned image, by sensing left and right edges, preferably in coincidence in all three color planes--or otherwise to determine edges in black. Such edge sensing is preferably done by convolution of the input pixel stream with a windowed vertical jump, yielding narrow peaks of opposite directedness at beginning and end of each character stroke. Textual material is indicated where such peaks alternate in directedness. Concurrent testing is desirable also for dark pixels between beginning and end of each stroke--and light pixels between end of one stroke and beginning of the next. In addition it is preferable to test for an appropriate (black and white) mix of pixels in each candidate text region. A preliminary test is conducted to determine whether a region being scanned is pictorial; in this case the entire edge-detection protocol with associated tests is bypassed, on the operating assumption that black-and-white text cannot be present in a region that is pictorial.

Abstract: In acquiring and reproducing an image, my method and apparatus correct for spurious gray background. Image brightness is sensed in multiple regions of an original image, to provide a corresponding multiplicity of signals for use in reproducing the image. In response to the multiple signals, a corresponding multiplicity of corrected signals is provided by adjusting each signal. The amount of adjustment depends on the size of the signal, and preferably includes jitter--i. e., a random component. Preferably the random component is partially smoothed. The multiple corrected signals are used in reproduction of the image. Preferably the dependence of adjustment on signal size is a multivalued monotonic function. Preferably the function increases the brightness of image regions that are already relatively more bright, in comparison with image regions that are relatively less bright.