Vision Framework ile Goruntu Isleme Rehberi

1import Vision
2import UIKit
3 
4// Temel Vision pipeline
5class VisionProcessor {
6 
7    func processImage(_ image: UIImage, requests: [VNRequest]) throws {
8        guard let cgImage = image.cgImage else {
9            throw VisionError.invalidImage
10        }
11 
12        // Handler olustur
13        let handler = VNImageRequestHandler(
14            cgImage: cgImage,
15            orientation: .up,
16            options: [:]
17        )
18 
19        // Request'leri calistir
20        try handler.perform(requests)
21    }
22 
23    // Video akisi icin sequence handler
24    func processVideoFrame(
25        _ pixelBuffer: CVPixelBuffer,
26        requests: [VNRequest]
27    ) throws {
28        let handler = VNImageRequestHandler(
29            cvPixelBuffer: pixelBuffer,
30            orientation: .up,
31            options: [:]
32        )
33 
34        try handler.perform(requests)
35    }
36}
37 
38enum VisionError: Error {
39    case invalidImage
40    case processingFailed
41}

1import Vision
2 
3class FaceDetector {
4 
5    func detectFaces(in image: CGImage) async throws -> [FaceResult] {
6        let request = VNDetectFaceRectanglesRequest()
7        request.revision = VNDetectFaceRectanglesRequestRevision3
8 
9        let handler = VNImageRequestHandler(cgImage: image)
10        try handler.perform([request])
11 
12        guard let observations = request.results else {
13            return []
14        }
15 
16        return observations.map { observation in
17            FaceResult(
18                boundingBox: observation.boundingBox,
19                confidence: observation.confidence,
20                roll: observation.roll?.doubleValue,
21                yaw: observation.yaw?.doubleValue,
22                pitch: observation.pitch?.doubleValue
23            )
24        }
25    }
26 
27    func detectFaceLandmarks(in image: CGImage) async throws -> [VNFaceObservation] {
28        let request = VNDetectFaceLandmarksRequest()
29        request.revision = VNDetectFaceLandmarksRequestRevision3
30 
31        let handler = VNImageRequestHandler(cgImage: image)
32        try handler.perform([request])
33 
34        return request.results ?? []
35    }
36}
37 
38struct FaceResult {
39    let boundingBox: CGRect
40    let confidence: Float
41    let roll: Double?
42    let yaw: Double?
43    let pitch: Double?
44}

1import Vision
2 
3class TextRecognizer {
4 
5    func recognizeText(
6        in image: CGImage,
7        languages: [String] = ["tr", "en"],
8        level: VNRequestTextRecognitionLevel = .accurate
9    ) async throws -> [RecognizedText] {
10        let request = VNRecognizeTextRequest()
11        request.recognitionLevel = level
12        request.recognitionLanguages = languages
13        request.usesLanguageCorrection = true
14 
15        let handler = VNImageRequestHandler(cgImage: image)
16        try handler.perform([request])
17 
18        guard let observations = request.results else {
19            return []
20        }
21 
22        return observations.compactMap { observation in
23            guard let candidate = observation.topCandidates(1).first else {
24                return nil
25            }
26 
27            return RecognizedText(
28                text: candidate.string,
29                confidence: candidate.confidence,
30                boundingBox: observation.boundingBox
31            )
32        }
33    }
34}
35 
36struct RecognizedText {
37    let text: String
38    let confidence: Float
39    let boundingBox: CGRect
40}

1import Vision
2 
3func detectBarcodes(in image: CGImage) throws -> [VNBarcodeObservation] {
4    let request = VNDetectBarcodesRequest()
5    request.symbologies = [
6        .qr, .ean13, .ean8, .code128, .code39, .upce,
7        .pdf417, .aztec, .dataMatrix
8    ]
9 
10    let handler = VNImageRequestHandler(cgImage: image)
11    try handler.perform([request])
12 
13    return request.results ?? []
14}

1import Vision
2import CoreML
3 
4class ImageClassifier {
5 
6    private lazy var classificationRequest: VNCoreMLRequest? = {
7        guard let model = try? VNCoreMLModel(
8            for: MobileNetV2(configuration: .init()).model
9        ) else { return nil }
10 
11        let request = VNCoreMLRequest(model: model) { request, error in
12            guard let results = request.results as? [VNClassificationObservation] else { return }
13 
14            let topResults = results.prefix(3)
15            for result in topResults {
16                print("\(result.identifier): %\(Int(result.confidence * 100))")
17            }
18        }
19        request.imageCropAndScaleOption = .centerCrop
20        return request
21    }()
22 
23    func classify(image: CGImage) throws {
24        guard let request = classificationRequest else { return }
25        let handler = VNImageRequestHandler(cgImage: image, options: [:])
26        try handler.perform([request])
27    }
28}